https://videos.concepttocloud.com/v/guardrails-your-ai-cant-talk-its-way-past Guardrails your AI cant talk its way past Free access plus LLM-written SQL might give decent results, but not every time. Scoped permissions, approvals and limits, enforced by the system rather than the prompt. Full video: https://youtu.be/JezW4GI1v04 Shorts AI agents datagovernance https://video-meta.open.video/poster/y-CWAbmRtD35/AsoAbVU5iR2_FeQjxF.jpg https://streaming.open.video/contents/y-CWAbmRtD35/1790809495/index.m3u8 https://videos.concepttocloud.com/embed?contentId=AsoAbVU5iR2 46 https://videos.concepttocloud.com/v/where-ai-guardrails-actually-belong Where AI guardrails actually belong Where do AI guardrails belong? In the data model: PII-safe pipelines that keep sensitive data inside the boundary, and decision logging on by default. Full video: https://youtu.be/7rcCUeXe1ZM Shorts AI dataprivacy compliance https://video-meta.open.video/poster/6pr8Eh8R7nLY/AYUBaUUPiQ2_sIjGbO.jpg https://streaming.open.video/contents/6pr8Eh8R7nLY/1790809621/index.m3u8 https://videos.concepttocloud.com/embed?contentId=AYUBaUUPiQ2 26 https://videos.concepttocloud.com/v/dont-just-give-your-llm-the-database Dont just give your LLM the database The easy way to put AI over your data: give an LLM the database and say have at it. Sometimes it infers the schema right. Sometimes is the problem. Full video: https://youtu.be/JezW4GI1v04 Shorts AI LLM agents https://video-meta.open.video/poster/-FvicxSJFvoA/AIUkH-UiPR2_GeqqoY.jpg https://streaming.open.video/contents/-FvicxSJFvoA/1790809622/index.m3u8 https://videos.concepttocloud.com/embed?contentId=AIUkH-UiPR2 44 https://videos.concepttocloud.com/v/wire-an-agent-to-real-data-without-it-going-rogue Wire an Agent to Real Data Without It Going Rogue An agent that cant touch anything is useless. An agent that can touch everything is dangerous. The answer is in the middle, and it isnt a better prompt. The easy way to put AI over your data is to hand an LLM the database and say have at it. Sometimes the SQL it writes is right. The problem is sometimes. In this one I walk through the alternative: a governed layer over your data, with scoped permissions, approvals and limits enforced by the system, so the same question gets the same, correct answer every time. Building with agents? https://concepttocloud.com/services/ai-data-preparation/ai-readiness-audit - The tension: useful access versus safe access - Why free access and LLM-written SQL isnt good enough - Scoped permissions, approvals and limits enforced by the system - Semantic models (Apache OSI, Snowflake, Malloy) so a metric means one thing 0:00 Useful and safe at the same time 0:42 How do I deploy AI over my data? 1:26 Should everyone LLM their own reports? 2:04 The wrong way: free access, LLM-written SQL 2:38 The right way: scoped permissions, approvals and limits 3:32 The security cost: training on and leaking your data 3:54 Most leaks start inside the building 4:34 Semantic models: teach the LLM how your data fits together 5:10 Gross profit, calculated two ways 5:49 Sensible security defaults, not lockdown Concept To Cloud builds agentic AI that is safe to run in production. Run the AI-Readiness Audit: https://concepttocloud.com/services/ai-data-preparation/ai-readiness-audit AI agents MCP aisafety https://video-meta.open.video/poster/8lzGArjJAfh6/AsURaVoiP72_vGCqTa.jpg https://streaming.open.video/contents/8lzGArjJAfh6/1790809639/index.m3u8 https://videos.concepttocloud.com/embed?contentId=AsURaVoiP72 415 https://videos.concepttocloud.com/v/guardrails-in-the-prompt-arent-guardrails Guardrails in the prompt arent guardrails Writing guardrails into the prompt doesnt hurt, but it isnt a control. Tell an auditor thats how you stop the model misbehaving and theyll send you back for a better answer. Full video: https://youtu.be/7rcCUeXe1ZM Shorts AI LLM compliance https://video-meta.open.video/poster/8Bf0ltT6Ffo6/6solb-VzOQ2_kyZTbe.jpg https://streaming.open.video/contents/8Bf0ltT6Ffo6/1790809622/index.m3u8 https://videos.concepttocloud.com/embed?contentId=6solb-VzOQ2 35 https://videos.concepttocloud.com/v/how-do-you-cut-into-a-system-nobody-understands How Do You Cut Into a System Nobody Understands? Can you use safe seams how do you cut into a system that no one fully understands without breaking it? That question drives everything else in a modernisation project. Full talk https://youtu.be/Tge0qCkKUn4 softwarearchitecture legacymodernisation engineering shorts https://video-meta.open.video/poster/z1qKgnetxnTA/7ZU6aooyP72_bQtmWU.jpg https://streaming.open.video/contents/z1qKgnetxnTA/1790809990/index.m3u8 https://videos.concepttocloud.com/embed?contentId=7ZU6aooyP72 12 https://videos.concepttocloud.com/v/the-strangler-fig-approach-to-legacy-rewrites The Strangler-Fig Approach to Legacy Rewrites The strangler fig approach: can you take a small part, modernize it, deploy it, get it running in production, and then move on to the next one? Replace it piece by piece, not all at once. Full talk https://youtu.be/Tge0qCkKUn4 stranglerfig softwarearchitecture legacymodernisation shorts https://video-meta.open.video/poster/8ImitlvcsfoY/BZUkbVpjOR2_OkOARs.jpg https://streaming.open.video/contents/8ImitlvcsfoY/1790809989/index.m3u8 https://videos.concepttocloud.com/embed?contentId=BZUkbVpjOR2 14 https://videos.concepttocloud.com/v/one-free-fix-for-your-whole-supply-chain One Free Fix for Your Whole Supply Chain If you take one operational change away from this: pin deployments to digests, not tags. Its free, its an afternoon of work, and it makes everything else in a provable supply chain possible. Full talk https://youtu.be/9B7TNdiCWuU devsecops softwaresupplychain sigstore shorts https://video-meta.open.video/poster/D2e8tz0BZDhz/7tURGEpjPQ2_EhRTHG.jpg https://streaming.open.video/contents/D2e8tz0BZDhz/1790809995/index.m3u8 https://videos.concepttocloud.com/embed?contentId=7tURGEpjPQ2 13 https://videos.concepttocloud.com/v/four-or-five-databases-when-one-would-do Four or Five Databases When One Would Do A lot of teams run four or five databases when one would do. Let me make the case for keeping it boring. Full video https://youtu.be/rLCGAf2JGj4 postgres postgresql database dataengineering shorts https://video-meta.open.video/poster/zYmall1cYnn5/7tpkbFojO72_qYcATv.jpg https://streaming.open.video/contents/zYmall1cYnn5/1790809988/index.m3u8 https://videos.concepttocloud.com/embed?contentId=7tpkbFojO72 10 https://videos.concepttocloud.com/v/ask-your-platform-where-the-test-set-is Ask Your Platform Where the Test Set Is If you take one thing from this: go and ask your platform where your best models test set is, and see how long it takes to find that out. Full talk https://youtu.be/v0lUudGbwQ mlops dataengineering machinelearning dataprovenance shorts https://video-meta.open.video/poster/B2iakDqAtbMY/BJplbooyP62_tNEMeu.jpg https://streaming.open.video/contents/B2iakDqAtbMY/1790809988/index.m3u8 https://videos.concepttocloud.com/embed?contentId=BJplbooyP62 9 https://videos.concepttocloud.com/v/it-works-isnt-the-finish-line-for-regulated-ai It works isnt the finish line for regulated AI In a regulated business, the finish line isnt it works. Its it survives an audit. And if the guardrail isnt in the data model, it doesnt exist. Full video: https://youtu.be/7rcCUeXe1ZM Shorts AI compliance regulatedAI https://video-meta.open.video/poster/z3vmBjudYbD6/Y6pkGUViPl2_XAaJqW.jpg https://streaming.open.video/contents/z3vmBjudYbD6/1790810116/index.m3u8 https://videos.concepttocloud.com/embed?contentId=Y6pkGUViPl2 17 https://videos.concepttocloud.com/v/a-very-expensive-compute-bill A Very Expensive Compute Bill If you use Delta Live tables as your underlying compute and then serve a web app off the back of it, youre going to have some very sad customers and a very expensive compute bill. Use the right tool for the reasons it was actually built. Full video https://youtu.be/rLCGAf2JGj4 postgres dataengineering backend shorts https://video-meta.open.video/poster/-uaWxzXl6rMW/YkVBbpozOk2_LcKOmO.jpg https://streaming.open.video/contents/-uaWxzXl6rMW/1790810118/index.m3u8 https://videos.concepttocloud.com/embed?contentId=YkVBbpozOk2 16 https://videos.concepttocloud.com/v/a-good-assessment-can-tell-you-not-to-buy A Good Assessment Can Tell You Not to Buy A good AI readiness assessment can tell you not to buy. Our job isnt to say yes, go ahead with AI no matter what. It doesnt make a difference to us whether or not the project goes ahead but the honesty does matter. Full video https://youtu.be/1vrKRjJr2Jg ai aireadiness machinelearning shorts https://video-meta.open.video/poster/zua0xB0d6bK7/YQV7aVVzOA2_kpeHGX.jpg https://streaming.open.video/contents/zua0xB0d6bK7/1790810116/index.m3u8 https://videos.concepttocloud.com/embed?contentId=YQV7aVVzOA2 17 https://videos.concepttocloud.com/v/a-fractional-cto-right-for-some-a-waste-of-money-for-others A Fractional CTO: Right for Some, a Waste of Money for Others A fractional CTO is senior technical leadership without the full-time salary. For some companies thats exactly right. For others its a waste of time and money. Full video https://youtu.be/8evt9kOEeXM fractionalcto startups cto techleadership shorts https://video-meta.open.video/poster/43ziBFWAZn_W/YkU6G-pPOA2_ucMabZ.jpg https://streaming.open.video/contents/43ziBFWAZn_W/1790810116/index.m3u8 https://videos.concepttocloud.com/embed?contentId=YkU6G-pPOA2 18 https://videos.concepttocloud.com/v/what-happens-when-your-llm-goes-offline What Happens When Your LLM Goes Offline? If youre baking an LLM into a product, what happens when it goes offline? Because these things have a habit of going offline. Does your product continue to work? What do those failure modes look like, and what can you do to mitigate the failure of an LLM service? Free AI-Readiness Audit https://concepttocloud.com/services/ai-data-preparation/ai-readiness-audit AI MLOps reliability dataengineering shorts https://video-meta.open.video/poster/7uqGgpWdBbL6/YkUQaUU4OB2_BObkXz.jpg https://streaming.open.video/contents/7uqGgpWdBbL6/1790810118/index.m3u8 https://videos.concepttocloud.com/embed?contentId=YkUQaUU4OB2 17 https://videos.concepttocloud.com/v/a-chromebook-but-for-linux A Chromebook, but for Linux Chromebook reliability with a GNOME desktop underneath. Give your granddad one and hell have a tough time breaking it and if he does, you flash it and start again. Full build-along https://youtu.be/nhx27jZDCL4 Shorts Linux Bluefin GNOME https://video-meta.open.video/poster/74eSdjGc3DvW/YRVBGUVPOB2_MICzvx.jpg https://streaming.open.video/contents/74eSdjGc3DvW/1790810617/index.m3u8 https://videos.concepttocloud.com/embed?contentId=YRVBGUVPOB2 20 https://videos.concepttocloud.com/v/this-isnt-research-it-ships-today This Isnt Research It Ships Today Sigstore and SLSA are open source, free, hosted by the OpenSSF, and run on the CI you already have. Theres no product here you have to buy to get started. Full talk https://youtu.be/9B7TNdiCWuU opensource openssf devsecops shorts https://video-meta.open.video/poster/98yykFGlZjCy/ZAoka_ozjB2_jCDqPg.jpg https://streaming.open.video/contents/98yykFGlZjCy/1790810626/index.m3u8 https://videos.concepttocloud.com/embed?contentId=ZAoka_ozjB2 21 https://videos.concepttocloud.com/v/good-a-solution-as-any-other-out-there Good a Solution as Any Other Out There There are reasons not to use Postgres. But if you want flexibility, decent speed, maintenance, and something you can Google the answer for nine times out of ten, Postgres is as good a solution as any other out there. Full video https://youtu.be/rLCGAf2JGj4 postgres postgresql database dataengineering shorts https://video-meta.open.video/poster/C4qKcpGl3r37/Z6Ulb_UjiB2_iAjIdi.jpg https://streaming.open.video/contents/C4qKcpGl3r37/1790810617/index.m3u8 https://videos.concepttocloud.com/embed?contentId=Z6Ulb_UjiB2 21 https://videos.concepttocloud.com/v/why-a-docker-tag-can-lie-to-you Why a Docker Tag Can Lie to You A tag is a label like a sticky note on a warehouse shelf. Someone can move it overnight and the note still reads the same. A digest is the fingerprint of the contents themselves: it cant be moved, only matched. Full talk https://youtu.be/9B7TNdiCWuU docker supplychainsecurity devops shorts https://video-meta.open.video/poster/D4m0dDbs3v0X/tkpBGpozjA2_IcplLr.jpg https://streaming.open.video/contents/D4m0dDbs3v0X/1790810625/index.m3u8 https://videos.concepttocloud.com/embed?contentId=tkpBGpozjA2 18 https://videos.concepttocloud.com/v/does-this-even-need-ai Does This Even Need AI? Before you build it: does what youre trying to do with AI actually add value, or would it be easier to build it out a different way and skip the hosting, liability, and complexity that comes with AI? A lot of the time people still go the AI route anyway. At least be aware of the option. Full video https://youtu.be/1vrKRjJr2Jg ai machinelearning dataengineering shorts https://video-meta.open.video/poster/C8yaxFastbFr/tQVlaVVijB2_XuUDDD.jpg https://streaming.open.video/contents/C8yaxFastbFr/1790810617/index.m3u8 https://videos.concepttocloud.com/embed?contentId=tQVlaVVijB2 18 https://videos.concepttocloud.com/v/what-an-ai-vendor-never-mentions What an AI Vendor Never Mentions We dont work like a sales pitch. Heres the red flag: if you speak to an AI vendor, theres unlikely to be any discussion about your data quality or compliance. What comes out the other end is a recommendation to use their product. Full video https://youtu.be/1vrKRjJr2Jg ai aireadiness dataengineering shorts https://video-meta.open.video/poster/88aKxFqlYr_s/ZAVRb_oPPB2_lsWAuX.jpg https://streaming.open.video/contents/88aKxFqlYr_s/1790810619/index.m3u8 https://videos.concepttocloud.com/embed?contentId=ZAVRb_oPPB2 21 https://videos.concepttocloud.com/v/why-freezing-your-legacy-system-doesnt-work Why Freezing Your Legacy System Doesnt Work Dont ask me how I know, but quite often the old system doesnt fully go away legacy customers, people who cant migrate. So even if you freeze a product and restart from scratch, quite often youre going to become a cropper. Full talk https://youtu.be/Tge0qCkKUn4 legacymodernisation softwarearchitecture engineering shorts https://video-meta.open.video/poster/yVvGFlucsnSq/tkpRH_VOOl2_CRccIq.jpg https://streaming.open.video/contents/yVvGFlucsnSq/1790810614/index.m3u8 https://videos.concepttocloud.com/embed?contentId=tkpRH_VOOl2 21 https://videos.concepttocloud.com/v/the-first-10-technical-decisions The First 10 Technical Decisions Before your first technical hire, you need the first 10 decisions to be the right ones. What platform are you using? Build versus buy? What direction from a technology perspective? You dont need a huge technical background to get this right, you just need someone whos answered these questions before. Full video https://youtu.be/8evt9kOEeXM fractionalcto startups founders shorts https://video-meta.open.video/poster/64eewFXkZj9s/t7UAb-oiik2_dgPTFZ.jpg https://streaming.open.video/contents/64eewFXkZj9s/1790810618/index.m3u8 https://videos.concepttocloud.com/embed?contentId=t7UAb-oiik2 21 https://videos.concepttocloud.com/v/we-migrated-a-1992-nasa-c-program-to-aws We Migrated a 1992 NASA C Program to AWS When I was working at NASA, we were tasked with moving a 1990s C program into AWS. Basically all the staff had left there was no one left who understood it. Heres how we did it anyway. Full talk https://youtu.be/Tge0qCkKUn4 NASA legacymodernisation AWS softwarearchitecture shorts https://video-meta.open.video/poster/C_nKAjLswjeY/YkoBHFpijQ2_TlizVB.jpg https://streaming.open.video/contents/C_nKAjLswjeY/1790810615/index.m3u8 https://videos.concepttocloud.com/embed?contentId=YkoBHFpijQ2 27 https://videos.concepttocloud.com/v/what-a-signature-actually-proves What a Signature Actually Proves A signature answers exactly one question: who vouched for this exact content? Not the vendor. Not the project. Not the version. This content, byte for byte change one byte anywhere, and verification fails. Full talk https://youtu.be/9B7TNdiCWuU sigstore cosign devsecops shorts https://video-meta.open.video/poster/zOu0cFXc2zwt/skolHpVji72_SGUPxE.jpg https://streaming.open.video/contents/zOu0cFXc2zwt/1790810625/index.m3u8 https://videos.concepttocloud.com/embed?contentId=skolHpVji72 27 https://videos.concepttocloud.com/v/training-is-the-easy-part Training Is the Easy Part A pipeline that runs in a sandbox and a pipeline that runs where failure is not an option are not the same system. What changes is everything around the model. Storage, for one. In a sandbox the model is wherever the notebook wrote it. In production it needs an addressable archive with a versioning policy you have to be able to name a thing and get exactly that thing back. Years later. Full talk https://youtu.be/v0lUudGbwQ mlops machinelearning dataengineering kubernetes shorts https://video-meta.open.video/poster/AVfqEp1Asrh7/s6p6bFpzi72_fRRZsc.jpg https://streaming.open.video/contents/AVfqEp1Asrh7/1790810621/index.m3u8 https://videos.concepttocloud.com/embed?contentId=s6p6bFpzi72 24 https://videos.concepttocloud.com/v/watch-it-update-while-i-look-away Watch it update while I look away Notes appearing on the wall in real time while I am looking at a completely different screen, because Supabase realtime is pushing them straight through Postgres. No polling, no refresh button, no extra service to run. Full build-along: https://youtu.be/got09pWBMG0 Shorts supabase postgres realtime webdev https://video-meta.open.video/poster/ySeOwnHc2r2Y/YkV6a_Vyi72_mZPgjm.jpg https://streaming.open.video/contents/ySeOwnHc2r2Y/1790810625/index.m3u8 https://videos.concepttocloud.com/embed?contentId=YkV6a_Vyi72 25 https://videos.concepttocloud.com/v/at-what-point-did-civilization-shut-down-for-two-months-to-upgrade-the-internet At what point did civilization shut down for two months to upgrade the internet? Dial-up to gigabit in 30 years. The internet never went offline for two months to do it. Thats the bar for a rebuild. If customers depend on your service, you cant change the interface and expect them to go read the docs. Nobody reads the docs. The upgrade has to happen underneath them. Ep.01 of the Concept To Cloud podcast. Full episode at the link in bio. productdevelopment softwarerebuild legacysystems cto productmanagement https://video-meta.open.video/poster/4_vOFBLlYjmZ/s6p7aEUiiR2_dnSzsb.jpg https://streaming.open.video/contents/4_vOFBLlYjmZ/1790810617/index.m3u8 https://videos.concepttocloud.com/embed?contentId=s6p7aEUiiR2 25 https://videos.concepttocloud.com/v/the-desktop-that-cant-drift The desktop that cant drift Production solved configuration drift a decade ago: you build an image, you ship it, and if its wrong you get the old one back. Bluefin points the same idea at a laptop the OS is an OCI image, so the machine cant drift because theres nothing to drift into. Full build-along https://youtu.be/nhx27jZDCL4 Shorts Linux Bluefin CloudNative https://video-meta.open.video/poster/B7uSEnKd7vp7/Y7VRGppOiR2_bPNQEF.jpg https://streaming.open.video/contents/B7uSEnKd7vp7/1790810878/index.m3u8 https://videos.concepttocloud.com/embed?contentId=Y7VRGppOiR2 29 https://videos.concepttocloud.com/v/your-laptop-is-haunted Your laptop is haunted Install something to test it. Follow a Stack Overflow answer at midnight. Add a repo and forget about it. Two years later nothing is obviously broken, but the laptop is haunted and the only fix on offer is to wipe it and start again. Full build-along https://youtu.be/nhx27jZDCL4 Shorts Linux Bluefin DevOps https://video-meta.open.video/poster/74zGsFusAbS4/YBUkb_pzP62_PprjJy.jpg https://streaming.open.video/contents/74zGsFusAbS4/1790810878/index.m3u8 https://videos.concepttocloud.com/embed?contentId=YBUkb_pzP62 30 https://videos.concepttocloud.com/v/a-run-is-a-verb-a-product-is-a-noun A Run Is a Verb. A Product Is a Noun. Your platform remembers runs. Every one of those questions is about a product. A run is a verb something that happened at a point in time. A product is a noun something that exists and persists. Platforms get built by people optimising the verb. Auditors, incident reporters and your future self all ask about the noun. Full talk https://youtu.be/v0lUudGbwQ mlops dataengineering kubeflow dataprovenance shorts https://video-meta.open.video/poster/FOjmAzesAbo4/s7Ukb-UyO62_vVdWPQ.jpg https://streaming.open.video/contents/FOjmAzesAbo4/1790810879/index.m3u8 https://videos.concepttocloud.com/embed?contentId=s7Ukb-UyO62 28 https://videos.concepttocloud.com/v/there-is-bias-in-product-feedback There is bias in product feedback Feature requests can only reflect the vocabulary of the product users already see. Which means your roadmap gets shaped by whats visible, not whats actually needed. Users dont ask for things they cant imagine inside the current interface. So the backlog fills up with incremental improvements to screens that might not even be the right screens. The fix isnt more feedback. Its watching workflows around the product, not inside it. Where do people drop out? Where do they open a spreadsheet because the tool didnt do the thing? This is what our Assess phase actually looks like in practice. One to two weeks going through the platform, the architecture, and how the team works, watching real workflows, not just reading code. We come back with a written report of whats blocking you and a costed plan to fix it. That plan belongs to you, no obligation to come back. Full episode on Concept to Cloud EP. 01 concepttocloud.com/podcast/concept-to-cloud/tack-on-or-rebuild productengineering productmanagement feedbackloops productstrategy softwareengineering https://video-meta.open.video/poster/9SzWwjXl_nU6/Y7UBHVozPQ2_JlkhXd.jpg https://streaming.open.video/contents/9SzWwjXl_nU6/1790810878/index.m3u8 https://videos.concepttocloud.com/embed?contentId=Y7UBHVozPQ2 27 https://videos.concepttocloud.com/v/nasas-mars-data-took-30-hours-we-got-it-to-10 NASAs Mars Data Took 30 Hours. We Got It to 10. A scientist could form a hypothesis on Mars and not get the answer back the same sol. NASAs PIXL instrument reads the chemistry of Martian rock live from the Perseverance rover. Processing that data took 30 hours long enough that the science had to wait for the pipeline. We rebuilt the cloud backend and got it to 10 minutes. 180x. Twelve weeks concept to production. NASA Software of the Year runner-up, 2023. Full story https://youtu.be/APhsIuB4QpY Work with the team that did this https://concepttocloud.com/contact NASA Mars dataengineering cloudengineering Perseverance shorts https://video-meta.open.video/poster/6SfyczGt7nL4/YBURHUVjOQ2_Ivbedb.jpg https://streaming.open.video/contents/6SfyczGt7nL4/1790810877/index.m3u8 https://videos.concepttocloud.com/embed?contentId=YBURHUVjOQ2 27 https://videos.concepttocloud.com/v/i-asked-claude-to-build-something-on-supabase I asked Claude to build something on Supabase I gave Claude Code one line: build something interesting on this Supabase project. It came back with a semantic sticky wall. A live notes wall with embeddings generated entirely locally by the edge runtimes built in GTE-small model, so semantic search runs with no API key and nothing leaving the machine. Auth, Postgres, RLS, pgvector, realtime, storage, edge functions and pgcron, all exercised in one app. Full build-along: https://youtu.be/got09pWBMG0 Shorts supabase claudecode pgvector ai https://video-meta.open.video/poster/8xmyclqY3vMY/ZkVkGpV4PR2_IydPzV.jpg https://streaming.open.video/contents/8xmyclqY3vMY/1790811014/index.m3u8 https://videos.concepttocloud.com/embed?contentId=ZkVkGpV4PR2 30 https://videos.concepttocloud.com/v/an-ai-wrote-this-manifest-it-broke-7-policies An AI Wrote This Manifest. It Broke 7 Policies. An AI coding assistant was asked for a Kubernetes manifest for a debug sidecar that could inspect node-level networking. This is what came back, unedited. It is confident. It is well formatted. It has a helpful comment next to the privileged flag explaining that it is needed for packet capture which is true. Every individual decision in it is defensible. Nobody was careless. It breaks seven separate policies. The interesting part is not that a machine wrote it. It is that it arrives in your cluster at three in the morning through a pipeline, with nobody in the loop to feel uneasy about it. Full talk policy as code on two real clusters, OPA Gatekeeper vs Kyverno: https://youtu.be/CzBgg8wedM kubernetes policyascode devsecops platformengineering shorts https://video-meta.open.video/poster/8veetpL62zgz/Z6V7bVo5OQ2_SIJInQ.jpg https://streaming.open.video/contents/8veetpL62zgz/1790811007/index.m3u8 https://videos.concepttocloud.com/embed?contentId=Z6V7bVo5OQ2 32 https://videos.concepttocloud.com/v/postgres-quietly-does-a-lot-more-than-select Postgres Quietly Does a Lot More Than SELECT You can use JSON out the box. You can bring in full text indexes for search. Youve got queues, PG vector, analytics, and a whole array of different extensions. Postgres quietly does a lot more than a select statement. Full video https://youtu.be/rLCGAf2JGj4 postgres postgresql dataengineering backend shorts https://video-meta.open.video/poster/zvaCBlaJZfmY/tRokbEoji72_Jvyist.jpg https://streaming.open.video/contents/zvaCBlaJZfmY/1790811010/index.m3u8 https://videos.concepttocloud.com/embed?contentId=tRokbEoji72 32 https://videos.concepttocloud.com/v/not-everybody-should-have-a-fractional-cto Not Everybody Should Have a Fractional CTO Not everybody should have a fractional CTO. If you have a strong lead who just needs a backup, hire a senior engineer, not a CTO. If the work is one defined project, scope the build and skip the leadership role entirely. Full video https://youtu.be/8evt9kOEeXM fractionalcto startups cto techleadership shorts https://video-meta.open.video/poster/CbumhBfR2rh7/t7UAbpUziQ2_MVFtAG.jpg https://streaming.open.video/contents/CbumhBfR2rh7/1790811006/index.m3u8 https://videos.concepttocloud.com/embed?contentId=t7UAbpUziQ2 29 https://videos.concepttocloud.com/v/most-companies-buy-the-ai-tool-first-thats-backwards Most Companies Buy the AI Tool First (Thats Backwards) Most companies buy the AI tool first and ask whether they were ready second. Thats backwards, and its expensive. An AI readiness assessment answers the question before you spend the money. Full video https://youtu.be/1vrKRjJr2Jg ai aireadiness machinelearning dataengineering shorts https://video-meta.open.video/poster/8rqKBB1ZwbVt/Z7olbFpii62_IXzonf.jpg https://streaming.open.video/contents/8rqKBB1ZwbVt/1790811013/index.m3u8 https://videos.concepttocloud.com/embed?contentId=Z7olbFpii62 31 https://videos.concepttocloud.com/v/embeddings-on-your-own-machine Embeddings on your own machine The embeddings are GTE-small running inside the Supabase edge runtime. On my laptop. No API key, no external calls, no egress. Semantic search with nothing confidential leaving the box is a very different conversation to have with a security reviewer. Full build-along: https://youtu.be/got09pWBMG0 Shorts supabase ai embeddings privacy https://video-meta.open.video/poster/EbiaEBH62nVX/tRo7a-UyP72_fkQDLJ.jpg https://streaming.open.video/contents/EbiaEBH62nVX/1790811014/index.m3u8 https://videos.concepttocloud.com/embed?contentId=tRo7a-UyP72 33 https://videos.concepttocloud.com/v/right-for-delivery-wrong-for-ownership Right for Delivery, Wrong for Ownership An agency is right for delivery, not for ownership. Theyll build your software on time and on budget. What they wont do is the leadership work: the judgment calls, the roadmap, the decisions that shape what gets built next. Thats a different problem. Full video https://youtu.be/8evt9kOEeXM fractionalcto startups cto shorts https://video-meta.open.video/poster/Aby0hzb63zKW/Z7VAGUozPQ2_INSedS.jpg https://streaming.open.video/contents/Aby0hzb63zKW/1790811007/index.m3u8 https://videos.concepttocloud.com/embed?contentId=Z7VAGUozPQ2 31 https://videos.concepttocloud.com/v/2212-ai-papers-zero-usable-models 2,212 AI Papers. Zero Usable Models. A 2021 Nature Machine Intelligence review screened 2,212 COVID-19 diagnosis and prognosis papers. 62 survived quality screening. None were judged of potential clinical use. The failures worth caring about are provenance failures: assembled datasets with the same images in train and test, and no record of where any of it came from. Full talk https://youtu.be/v0lUudGbwQ kubeflow mlops dataengineering machinelearning shorts https://video-meta.open.video/poster/5rmeljuRwjS5/tRVBaoozO62_ESJgiq.jpg https://streaming.open.video/contents/5rmeljuRwjS5/1790811010/index.m3u8 https://videos.concepttocloud.com/embed?contentId=tRVBaoozO62 33 https://videos.concepttocloud.com/v/80-of-ai-projects-end-up-on-the-scrap-heap 80% of AI Projects End Up on the Scrap Heap A couple of years ago a review put it at 95% of AI projects ending up on the scrap heap. One from earlier this year had it down to about 80%. Thats still a lot of projects going through the motions. The demos, the testing, the ideas all seem sound and then they dont actually ship. Free AI-Readiness Audit https://concepttocloud.com/services/ai-data-preparation/ai-readiness-audit AI machinelearning dataengineering MLOps shorts https://video-meta.open.video/poster/DbqCBnKZtbLq/tBVBbEVzPR2_lbbqBl.jpg https://streaming.open.video/contents/DbqCBnKZtbLq/1790811010/index.m3u8 https://videos.concepttocloud.com/embed?contentId=tBVBbEVzPR2 33 https://videos.concepttocloud.com/v/talking-to-your-data-claude-mcp-from-excel-to-product-ep-8 Talking to Your Data (Claude + MCP) From Excel to Product (Ep. 8) What if you could just ask? In episode 8 I wire Claude Code to Saikus MCP server listing cubes, querying drug sales and rebates in plain English, with row-level security still enforced. In this episode: Connecting Claude to the Saiku MCP server Querying drug sales & rebate % conversationally Why the model (not the raw DB) keeps answers safe Chapters: 0:00 Claude + the Saiku MCP server 5:12 Querying drug sales 12:44 Rebates analysis 15:15 Row-level results Concept to Cloud: https://concepttocloud.com Saiku (open source): https://github.com/spiculedata/saiku dataengineering analytics excel saiku dataproducts BI https://video-meta.open.video/poster/5JeekDa6tfuB/Z7UAHooPP62_guOcQN.jpg https://streaming.open.video/contents/5JeekDa6tfuB/1790811300/index.m3u8 https://videos.concepttocloud.com/embed?contentId=Z7UAHooPP62 1189 https://videos.concepttocloud.com/v/local-is-one-thing-production-is-another Local is one thing. Production is another Spin up Supabase locally and you get Postgres, auth, storage, edge functions and realtime in a bunch of Docker containers. That part is easy. Repeating it in a VM in the cloud is also easy. Dealing with it from a scale, maintenance and security perspective is the actual job, and that is a different job. Full build-along: https://youtu.be/got09pWBMG0 Shorts supabase postgres startups devops https://video-meta.open.video/poster/-syaslGZAb2s/6QoBGEUyOk2_fdsJmC.jpg https://streaming.open.video/contents/-syaslGZAb2s/1790811274/index.m3u8 https://videos.concepttocloud.com/embed?contentId=6QoBGEUyOk2 39 https://videos.concepttocloud.com/v/a-home-server-your-agents-can-drive A home server your agents can drive Bluefin Server is image-based, self-updating and organised around Kubernetes and it says its fully API and MCP driven. That means your own agents talking to your own home server. Full build-along https://youtu.be/nhx27jZDCL4 Shorts Linux Bluefin MCP AI https://video-meta.open.video/poster/-uaKEpvI_D3Z/6Qp6aooyPB2_qeDbAc.jpg https://streaming.open.video/contents/-uaKEpvI_D3Z/1790811270/index.m3u8 https://videos.concepttocloud.com/embed?contentId=6Qp6aooyPB2 36 https://videos.concepttocloud.com/v/does-your-policy-engine-fail-open-or-closed Does Your Policy Engine Fail Open or Closed? The question almost nobody asks until the audit: what happens to admission when the policy engine itself is down? Because it will be. It is a pod, and pods get evicted, upgraded, or fail to come back up. Gatekeeper ships failurePolicy: Ignore with a 3 second timeout the upstream default. If it is down, slow, or mid-upgrade, workloads are admitted and the audit sweep flags it a minute later, by which point it is running. Kyverno is the other way around. It fails closed, with a 10 second timeout. Neither default is wrong. Fail-open protects the cluster, fail-closed protects the policy. An admission webhook is a structural control only if it fails closed if it fails open, then for the duration of the outage it is an advisory control in a high-vis jacket. Go and check which one you installed. Most people never have. Full talk policy as code on two real clusters, OPA Gatekeeper vs Kyverno: https://youtu.be/CzBgg8wedM kubernetes policyascode sre platformengineering shorts https://video-meta.open.video/poster/6gqqdpLR6jCY/A6pkHVozPk2_NpWVYV.jpg https://streaming.open.video/contents/6gqqdpLR6jCY/1790811278/index.m3u8 https://videos.concepttocloud.com/embed?contentId=A6pkHVozPk2 39 https://videos.concepttocloud.com/v/a-prompt-is-a-wiki-that-talks-back A Prompt Is a Wiki That Talks Back If you are regulated, at some point somebody asks for evidence as a control. There is an enormous difference between we have a policy that says engineers shouldnt do this and here is the webhook that structurally prevents it, here is its configuration, and here is the admission log showing it firing. One of those is a conversation you have with an auditor. The other one is a finding. An advisory control can only ever be evidenced by asking people whether they followed it. Which is why it gets twitchy when someone says they put it in the prompt because a prompt is a wiki that talks back. Full talk policy as code on two real clusters, OPA Gatekeeper vs Kyverno: https://youtu.be/CzBgg8wedM kubernetes policyascode compliance devsecops shorts https://video-meta.open.video/poster/ygumwD06Aj0z/AkpRbopiPA2_NqWQrJ.jpg https://streaming.open.video/contents/ygumwD06Aj0z/1790811274/index.m3u8 https://videos.concepttocloud.com/embed?contentId=AkpRbopiPA2 40 https://videos.concepttocloud.com/v/quarterly-roadmaps-now-last-a-week-and-a-half Quarterly roadmaps now last a week and a half. Agentic engineers are churning through work so fast that code isnt the bottleneck anymore. Product is. If your product org cant keep up with what engineering can now ship, youve got the wrong constraint. Full episode concepttocloud.com/podcast/concept-to-cloud/tack-on-or-rebuild https://video-meta.open.video/poster/CuyGln0Q_bvr/A6pRb_UzOA2_IgAyjy.jpg https://streaming.open.video/contents/CuyGln0Q_bvr/1790811270/index.m3u8 https://videos.concepttocloud.com/embed?contentId=A6pRb_UzOA2 37 https://videos.concepttocloud.com/v/how-to-ship-ai-that-survives-a-compliance-audit How to Ship AI That Survives a Compliance Audit In a regulated business, an AI feature isnt done when it works its done when it survives an audit. Heres how to build AI where every decision leaves a paper trail and sensitive data never leaves the boundary. Check your AI is audit-ready: https://concepttocloud.com/services/ai-data-preparation/ai-readiness-audit - Why regulated AI is different audit trail, PII, explainability - The wrong approach: guardrails in the prompt - The right approach: guardrails in the data model, PII-safe pipelines, decision logging - What auditors actually ask and how to have the answer ready 0:00 In a regulated business, it works isnt the finish line 0:49 Why regulated AI is different 1:21 PII cant leak, and you have to explain the output 2:12 The wrong approach: guardrails in the prompt 3:03 The right approach: guardrails in the data model 3:19 PII-safe pipelines, decision logging on by default 4:21 What auditors actually ask 4:43 The hard one: can you reproduce it? 5:24 The rule Correction: at 4:11 I said mutable. I meant immutable the decision log should be something nobody can change after the fact. Concept To Cloud ships AI that survives audit in regulated industries. Run the AI-Readiness Audit: https://concepttocloud.com/services/ai-data-preparation/ai-readiness-audit AI compliance regulated MLOps dataprivacy https://video-meta.open.video/poster/7ceypBfQBz-z/6RolHUUyPk2_MrLLOD.jpg https://streaming.open.video/contents/7ceypBfQBz-z/1790811322/index.m3u8 https://videos.concepttocloud.com/embed?contentId=6RolHUUyPk2 367 https://videos.concepttocloud.com/v/would-i-ship-this-to-production-absolutely-not Would I Ship This to Production? Absolutely Not. Would I ship what we just built straight to production? Absolutely not. Security, scale, and a long list of caveats stand between a live demo and something a regulated team can actually run. That gap is the whole reason Concept To Cloud exists. Book a free AI-Readiness Audit https://concepttocloud.com AI DataPrivacy CloudEngineering shorts https://video-meta.open.video/poster/8gmKkjLJAjF6/6BpQb-piOB2_yQlFlx.jpg https://streaming.open.video/contents/8gmKkjLJAjF6/1790811312/index.m3u8 https://videos.concepttocloud.com/embed?contentId=6BpQb-piOB2 55 https://videos.concepttocloud.com/v/rags-missing-piece-whos-allowed-to-see-what RAGs Missing Piece: Whos Allowed to See What Retrieval that hands back everything is easy. Retrieval that only hands back what the person asking is actually allowed to see is the hard, necessary part. Row-level access and PII handling are the two pieces most retrieval demos skip. Book a free AI-Readiness Audit https://concepttocloud.com DataPrivacy AI RowLevelSecurity shorts https://video-meta.open.video/poster/4qCqkDfQEDxZ/6RV7GpoOOA2_oeDESP.jpg https://streaming.open.video/contents/4qCqkDfQEDxZ/1790811271/index.m3u8 https://videos.concepttocloud.com/embed?contentId=6RV7GpoOOA2 41 https://videos.concepttocloud.com/v/48-hours-with-opus-55-is-it-actually-worth-it-shorts 48 Hours With Opus 5.5: Is It Actually Worth It? shorts 48 hours in with Opus 5.5. Juans honest take: one of the better models Anthropic has put out, 4.8 vibes with Fable logic, and cheaper too. He uses it for development and programming. What are you using it for? Drop it in the comments. Is your AI stack actually ready for real data and real production traffic? Concept To Clouds free AI-Readiness Audit tells you where the gaps are: https://concepttocloud.com/services/ai-data-preparation/ai-readiness-audit Concept To Cloud helps regulated and mid-size teams take AI from toy demo to production. concepttocloud.com shorts opus claude anthropic ai claudecode https://video-meta.open.video/poster/FsuShjWI-v9y/AlV7bEU5Pl2_ufDtOv.jpg https://streaming.open.video/contents/FsuShjWI-v9y/1790811278/index.m3u8 https://videos.concepttocloud.com/embed?contentId=AlV7bEU5Pl2 44 https://videos.concepttocloud.com/v/i-tried-to-break-my-own-ai-agent I Tried to Break My Own AI Agent Before we shipped guardrails, we tried to break our own agent asking it something completely off-topic just to see what it would do. It answered. Thats the failure mode every AI agent needs a fix for before it goes anywhere near production. Book a free AI-Readiness Audit https://concepttocloud.com AIAgents AI Guardrails shortsbeta https://video-meta.open.video/poster/ywaaoFr66z3z/7QVAHEVijB2_ZuQkPz.jpg https://streaming.open.video/contents/ywaaoFr66z3z/1790811292/index.m3u8 https://videos.concepttocloud.com/embed?contentId=7QVAHEVijB2 41 https://videos.concepttocloud.com/v/dont-spend-three-months-choosing-a-policy-engine Dont Spend Three Months Choosing a Policy Engine Reach for Gatekeeper and OPA when your policy problem is bigger than Kubernetes Terraform, CI, Envoy, application authz and you want one language across all of it. Reach for Kyverno when Kubernetes is the whole problem, when you want every engineer writing and reviewing policy rather than just the platform team, or when you need mutation, generation, or image provenance without standing up a second system. A lot of regulated shops land on both: Kyverno for the Kubernetes 80%, Rego kept for CI and infrastructure-as-code. But whichever you pick dont spend three months choosing. The rules matter more than the engine, they are portable between the two, and the cost of picking the less ideal one is far lower than the cost of another quarter with no admission control at all. Full talk policy as code on two real clusters, OPA Gatekeeper vs Kyverno: https://youtu.be/CzBgg8wedM kubernetes opa kyverno platformengineering shorts https://video-meta.open.video/poster/4giCBlXJ7D2W/BkVRbppijB2_bzqKJw.jpg https://streaming.open.video/contents/4giCBlXJ7D2W/1790811287/index.m3u8 https://videos.concepttocloud.com/embed?contentId=BkVRbppijB2 40 https://videos.concepttocloud.com/v/the-pii-redaction-reveal The PII Redaction Reveal Credit card, email, phone number all automatically redacted before they ever reach the model. Its not a perfect first pass, but its the second half of doing retrieval safely: keep sensitive fields out of what the AI actually sees. Book a free AI-Readiness Audit https://concepttocloud.com PIIRedaction DataPrivacy AI shorts https://video-meta.open.video/poster/7cyCclf77n8r/7QokbFp5iA2_rRyecr.jpg https://streaming.open.video/contents/7cyCclf77n8r/1790811297/index.m3u8 https://videos.concepttocloud.com/embed?contentId=7QokbFp5iA2 43 https://videos.concepttocloud.com/v/i-have-2-notes-the-wall-has-34 I Have 2 Notes. The Wall Has 34. I have 2 notes on this wall. The wall has 34 total. Thats row-level security working exactly as intended the chatbot only ever hands back what the person asking is actually allowed to see, no matter how much data sits behind it. Book a free AI-Readiness Audit https://concepttocloud.com RowLevelSecurity DataPrivacy Postgres shorts https://video-meta.open.video/poster/9kDSoBrRtDhy/7kV6Hoo4OA2_ZmMxfa.jpg https://streaming.open.video/contents/9kDSoBrRtDhy/1790811567/index.m3u8 https://videos.concepttocloud.com/embed?contentId=7kV6Hoo4OA2 45 https://videos.concepttocloud.com/v/what-happens-when-your-ai-provider-goes-down What Happens When Your AI Provider Goes Down? What happens to your product the day your model provider takes an afternoon off? Every team shipping an LLM-driven feature needs an answer to that question before it becomes an incident. This is the part of build an AI agent nobody puts in the demo. Book a free AI-Readiness Audit https://concepttocloud.com AIAgents AI Reliability shorts https://video-meta.open.video/poster/yofWwnL6sDoy/7QpAGVU5Ol2_UYttwW.jpg https://streaming.open.video/contents/yofWwnL6sDoy/1790811562/index.m3u8 https://videos.concepttocloud.com/embed?contentId=7QpAGVU5Ol2 47 https://videos.concepttocloud.com/v/what-an-ai-agent-actually-is What an AI Agent Actually Is An AI agent isnt magic its a model with tools, wired up to go do something specific with your real data and hand back a result. We built one live this week on top of a real dataset, not a toy demo. Book a free AI-Readiness Audit https://concepttocloud.com AIAgents AI MachineLearning shorts https://video-meta.open.video/poster/yEf0pDaIwDVz/7kpBbFUOOB2_nuBRDg.jpg https://streaming.open.video/contents/yEf0pDaIwDVz/1790811566/index.m3u8 https://videos.concepttocloud.com/embed?contentId=7kpBbFUOOB2 46 https://videos.concepttocloud.com/v/the-red-flag-that-surfaces-the-week-after-close The Red Flag That Surfaces the Week After Close Integration and vendor lock-in is red flag five. Its the one that shows up the week after close, once portfolio companies start getting pushed together. Reducing outlay is part of the point of private equity. So the diligence question is narrow: is there anything in this stack that blocks a cloud migration, or a data-service migration, that would otherwise take cost out? These arent always deal-breakers. They belong in the report anyway, because someone pays for them later. Pre-acquisition checklist https://concepttocloud.com/resources/pre-acquisition-tech-diligence-checklist privateequity duediligence vendorlockin cloudmigration shorts https://video-meta.open.video/poster/AybSFjv6tfpz/B7oAHppzjk2_wvnXqY.jpg https://streaming.open.video/contents/AybSFjv6tfpz/1790811562/index.m3u8 https://videos.concepttocloud.com/embed?contentId=B7oAHppzjk2 49 https://videos.concepttocloud.com/v/tech-due-dillgence-is-a-not-a-code-review-pe-duediligence Tech Due Dillgence Is A Not A Code Review pe duediligence Were not grading the code. Were pricing the gap between what they have and what your thesis needs. A technical due diligence answers one question: does the technology support the investment thesis and if not, what will it cost to get there? The layers we actually read: product & UX, architecture, data, security & compliance, team & key-person risk, and roadmap vs the plan. Full breakdown https://youtu.be/AZcfGJ8jJGg Free Tech-Risk Scorecard https://concepttocloud.com/services/technical-due-diligence/tech-risk-scorecard privateequity duediligence techdiligence MandA privateequityinvesting shorts https://video-meta.open.video/poster/5Ab0opf7YbMz/7Bo6GUUzjB2_dcHzXa.jpg https://streaming.open.video/contents/5Ab0opf7YbMz/1790811566/index.m3u8 https://videos.concepttocloud.com/embed?contentId=7Bo6GUUzjB2 47 https://videos.concepttocloud.com/v/prove-your-software-supply-chain-sigstore-slsa-in-practice Prove Your Software Supply Chain: Sigstore + SLSA in Practice Live engineering deep dive from Concept to Cloud. We make a software supply chain provable Sigstore signing, SLSA provenance, and verify-at-deploy for teams that answer to auditors. Live Q&A. https://concepttocloud.com https://video-meta.open.video/poster/z_u8BlKQxjLW/7BolbEoOjl2_WgTuHj.jpg https://streaming.open.video/contents/z_u8BlKQxjLW/1790812143/index.m3u8 https://videos.concepttocloud.com/embed?contentId=7BolbEoOjl2 3498 https://videos.concepttocloud.com/v/i-tried-to-beg-my-ai-agent-into-breaking-its-rules I Tried to Beg My AI Agent Into Breaking Its Rules After adding guardrails, we tried to talk our own agent out of them begging it to answer a question it had been told to refuse. It held the line. A deterministic gate, not an LLMs judgment call, decided what got answered. Book a free AI-Readiness Audit https://concepttocloud.com AIAgents AI Guardrails shorts https://video-meta.open.video/poster/4knKgpuZ2vgy/AQVQbVojjR2_OuBFKC.jpg https://streaming.open.video/contents/4knKgpuZ2vgy/1790811563/index.m3u8 https://videos.concepttocloud.com/embed?contentId=AQVQbVojjR2 55 https://videos.concepttocloud.com/v/build-a-product-end-to-end-with-free-ai-tools-ai-product-development-freeaitools Build a product end to end with free AI tools ai product development freeaitools Ship a software product end to end without paying for a subscription. opencode a free, open, terminal-based AI coding environment connected to OpenRouter for model access. Set your OpenRouter API key inside opencode, pick a model (Nvidias Nemotron is one free option), and define your dev agents by the task you actually need them to do. Full guide on which models to use for which tasks https://concepttocloud.com/free-tools/ai-tools https://concepttocloud.com ai freeaitools opencode openrouter softwaredevelopment shorts https://video-meta.open.video/poster/CEeudFH6_nLq/A6p6aVVji62_gUPlnx.jpg https://streaming.open.video/contents/CEeudFH6_nLq/1790811820/index.m3u8 https://videos.concepttocloud.com/embed?contentId=A6p6aVVji62 56 https://videos.concepttocloud.com/v/ai-theatre-in-due-diligence AI Theatre in Due Diligence When working on Technical Due Dilligence, what is AI Theatre? How do companies attempt to pull some smoke and mirrors and their AI product is really nothing more than a fancy proof of concept? How do you get from here to production? https://video-meta.open.video/poster/zEqysnXQ_bMt/A6pRbUVjjR2_qjozdz.jpg https://streaming.open.video/contents/zEqysnXQ_bMt/1790811820/index.m3u8 https://videos.concepttocloud.com/embed?contentId=A6pRbUVjjR2 70 https://videos.concepttocloud.com/v/is-your-ai-product-just-smoke-and-mirrors Is your AI product just smoke and mirrors? Could you do the same thing with a script and the data rather than an LLM and a bunch of tokens? Every deal has an AI story now. The question is whether that AI project is viable, whether it adds real value, and whether it drives customer growth or whether its a POC that will never ship. That answer changes what the equity is worth. Full breakdown https://youtu.be/AZcfGJ8jJGg Free Tech-Risk Scorecard https://concepttocloud.com/services/technical-due-diligence/tech-risk-scorecard privateequity duediligence AI techdiligence MandA shorts https://video-meta.open.video/poster/AAiConHI7vKZ/6QoAG_o5jR2_Ffcixm.jpg https://streaming.open.video/contents/AAiConHI7vKZ/1790811824/index.m3u8 https://videos.concepttocloud.com/embed?contentId=6QoAG_o5jR2 49 https://videos.concepttocloud.com/v/kubeflow-in-practice-ml-pipelines-on-kubernetes Kubeflow in Practice: ML Pipelines on Kubernetes Live engineering deep dive from Concept to Cloud. A hands-on look at Kubeflow for running real ML pipelines on Kubernetes, plus an honest take on when the Kubernetes tax is worth it. Live Q&A. https://concepttocloud.com https://video-meta.open.video/poster/74uSElmdEzv5/YZNdOopOPB2_SlfDsK.jpg https://streaming.open.video/contents/74uSElmdEzv5/1790813588/index.m3u8 https://videos.concepttocloud.com/embed?contentId=YZNdOopOPB2 3631 https://videos.concepttocloud.com/v/your-ai-project-is-really-a-data-project Your AI Project Is Really a Data Project Almost every AI project were handed is really a data project wearing a costume. Teams pour the excitement into the model and treat the data as a second-class citizen and thats why so many AI proof-of-concepts never leave proof-of-concept. Garbage in, garbage out still prevails in the world of LLMs. And garbage in, expensive cost out is very real, because token-based spend is bursty and you cant forecast it once a feature is client-facing. The fix is underneath: a semantic layer over your database, so everything is measured the same way and every surface LLMs, analysis, dashboards, other applications is built on one model. Is your data AI-ready? https://concepttocloud.com/services/ai-data-preparation/ai-readiness-audit - Why AI value comes from the data layer, not the model - Why POCs stall, and what changes when they hit production - The token-cost problem nobody models up front - What getting the data right actually means Chapters 0:00 Your AI project is a data project 0:43 Why most POCs never leave POC 1:09 The cost nobody models: tokens 1:49 POC economics arent production economics 2:52 So what does getting it right mean? 3:02 The semantic layer 3:28 One model, every surface 3:55 Take a step back 4:47 Get the layer right Concept To Cloud gets your data AI-ready so production AI is fast and cheap to build. Run the AI-Readiness Audit: https://concepttocloud.com/services/ai-data-preparation/ai-readiness-audit AI datastrategy semanticlayer dataengineering https://video-meta.open.video/poster/jAemoveaBhWX/ZlNJj-V4i72_IezugK.jpg https://streaming.open.video/contents/jAemoveaBhWX/1790815185/index.m3u8 https://videos.concepttocloud.com/embed?contentId=ZlNJj-V4i72 329 https://videos.concepttocloud.com/v/5-technical-red-flags-that-kill-acquisitions 5 Technical Red Flags That Kill Acquisitions Most deals dont die because the code is ugly. They die because of five specific technical risks that dont surface until its too late. Here they are - and how to spot them before you sign. The full pre-acquisition checklist: https://concepttocloud.com/resources/pre-acquisition-tech-diligence-checklist Five technical red flags, and why each one changes the deal: - Key-person / tribal knowledge risk - The hidden scalability ceiling - Security & compliance debt - AI theatre a roadmap of demos that never ship - Integration & vendor lock-in landmines Chapters 0:00 Its not the ugly code that kills deals 0:10 1 Key-person risk 1:06 2 Scalability ceiling 2:32 3 Security & compliance debt 4:15 4 AI theatre 5:26 5 Integration landmines 6:53 How we help get the checklist Concept to Cloud runs board-ready technical due diligence for private equity. Download the Diligence Checklist: https://concepttocloud.com/resources/pre-acquisition-tech-diligence-checklist privateequity duediligence M&A techdiligence dealmaking https://video-meta.open.video/poster/nRbeAtajAdIr/6AMcj_pzPl2_TzdXkh.jpg https://streaming.open.video/contents/nRbeAtajAdIr/1790815276/index.m3u8 https://videos.concepttocloud.com/embed?contentId=6AMcj_pzPl2 468 https://videos.concepttocloud.com/v/how-we-cut-nasas-mars-rover-data-pipeline-from-30-hours-to-10-minutes How We Cut NASAs Mars Rover Data Pipeline From 30 Hours to 10 Minutes NASAs PIXL instrument scans Martian rock chemistry live from the Perseverance rover - but processing took 30 hours, so a scientists hypothesis and its answer couldnt share the same day on Mars. We rebuilt the cloud backend and cut it to 10 minutes. Work with the team that did this: https://concepttocloud.com/contact The story behind Pixlise - NASA Software of the Year runner-up, 2023: - What PIXL does on the Perseverance rover, and why speed mattered - Why the bottleneck was the pipeline, not the science - How we re-architected the cloud backend - The result: 30 hours to 10 minutes (180x), concept to production in 12 weeks - What it means for your systems: your bottleneck is rarely the algorithm Chapters 0:00 A hypothesis and its answer, on Mars 0:24 Who we are 0:50 What PIXL does 1:12 You cant patch systems on Mars 1:43 Why 30 hours killed the science 2:16 The science was fine the pipeline wasnt 2:42 Re-architecting the backend 3:45 The platform: open source 4:07 30 hours to 10 minutes 4:45 The hard part: knowing you have it all 5:40 The lesson for your business 6:00 Work with us Concept to Cloud builds critical systems for missions that cant afford to fail ex-NASA engineers, senior-led. Book a discovery call: https://concepttocloud.com/contact NASA cloudengineering dataengineering Mars casestudy https://video-meta.open.video/poster/POCedt1z_BqA/AQgJOpoyOl2_leHXYz.jpg https://streaming.open.video/contents/POCedt1z_BqA/1790815272/index.m3u8 https://videos.concepttocloud.com/embed?contentId=AQgJOpoyOl2 392 https://videos.concepttocloud.com/v/postgres-is-probably-all-you-need-stop-adding-databases Postgres Is Probably All You Need (Stop Adding Databases) Most teams reach for a separate database for search, a queue, a cache, a vector store, and analytics. For a long time, Postgres can do all of it. I make the case for keeping your stack boring, when one database is genuinely enough, and the few times it is not. Not sure what your data actually needs? https://concepttocloud.com/contact What is in the video: Why teams over-collect databases What Postgres quietly does well: JSON, search, queues, vectors, analytics The real cost of every extra data store you run When you genuinely do need something else How to decide, without the hype Chapters 0:00 A lot of teams run four or five databases when one would do 0:36 Why teams over-collect databases 1:46 Resume-driven, hype-driven, cargo-culted 2:28 The database that stuck around 3:03 A pluggable architecture 3:53 JSON, full-text search, queues, pgvector, analytics 4:51 One platform, one maintenance area 5:38 Ops, failure modes, people 6:51 There is a time when you need more 7:30 Use Databricks or Snowflake for what theyre for 7:52 Talk it through, no pitch Concept to Cloud builds data systems sized to the problem, not the trend. Ex-NASA engineers, senior-led, fixed price. Book a discovery call https://concepttocloud.com/contact postgres postgresql database dataengineering backend https://video-meta.open.video/poster/SfDqBfGOYtJW/6BhdOEoOjl2_axveYF.jpg https://streaming.open.video/contents/SfDqBfGOYtJW/1790815413/index.m3u8 https://videos.concepttocloud.com/embed?contentId=6BhdOEoOjl2 535 https://videos.concepttocloud.com/v/when-to-hire-a-fractional-cto-and-when-you-really-should-not When to Hire a Fractional CTO (and When You Really Should Not) A fractional CTO is senior technical leadership without the full-time hire. Useful when you need the judgment but not the headcount. A waste of money in a few specific cases. Here is how to tell which one you are in. Talk it through, no pitch https://concepttocloud.com/contact I act as a fractional and embedded technical lead for founders and portfolio companies. In this video: What a fractional CTO actually does day to day The three moments it makes sense: pre-first-hire, post-funding, and mid-crisis When you should not hire one, and what to do instead Fractional vs full-time vs an agency, and the real cost of each How to work with one so you actually get value Chapters 0:00 What a fractional CTO is 0:35 Who this video is for 0:55 What a CTO actually does 2:08 Fractional: CTO oversight without the full-time cost 2:28 Moment 1: Pre-first-hire 3:27 Moment 2: Post-funding 4:05 Moment 3: Mid-crisis 4:51 When NOT to hire one 5:38 Full-time: right when the load is constant 6:01 Agency: right for delivery, wrong for ownership 6:34 Fractional: right when you need judgment, not hours 7:26 What to do next Concept to Cloud embeds senior, ex-NASA engineering leadership into founding teams and portfolio companies. Senior-led, fixed scope. Book a discovery call https://concepttocloud.com/contact fractionalcto startups cto techleadership founders https://video-meta.open.video/poster/Uvjutbb4spO7/6BNJioU5jl2_HvfuXA.jpg https://streaming.open.video/contents/Uvjutbb4spO7/1790815411/index.m3u8 https://videos.concepttocloud.com/embed?contentId=6BNJioU5jl2 505 https://videos.concepttocloud.com/v/what-an-ai-readiness-assessment-actually-checks-before-you-spend-a-dollar-on-ai What an AI Readiness Assessment Actually Checks (Before You Spend a Dollar on AI) Most companies buy the AI tool first and ask whether they were ready second. An AI readiness assessment tells you that in a week, not six months in. Here is what one actually checks, and how to spot a report that is just a sales pitch in disguise. Get a free AI-Readiness Audit https://concepttocloud.com/services/ai-data-preparation/ai-readiness-audit I run these for regulated and mid-size teams. In this video I break down what a real AI readiness assessment looks at: Data readiness: is your data clean, governed, and legal to use Security and compliance: what has to be true before sensitive data goes near a model The use case: whether the thing you want to automate is even a good fit for AI Team and process: who owns it after the pilot, and how it gets to production Cost and risk: the honest number, not the demo number Buy vs build vs wait Chapters 0:00 Why most AI projects fail the readiness test 0:33 Who this video is for 1:32 What an assessment actually is (not a demo) 1:56 Check 1: Data readiness 2:47 Check 2: Security & compliance 4:35 Check 3: Does this even need AI? 5:22 Check 4: Who owns it once its live? 6:09 Check 5: The path to production 6:55 Check 6: The real cost, not the demo cost 8:02 A good assessment can tell you not to buy 8:41 Red flag: no compliance view, thats a sales pitch 9:16 What it costs and where to start Free: the AI-Readiness Audit https://concepttocloud.com/services/ai-data-preparation/ai-readiness-audit Concept to Cloud builds and assesses production AI for regulated and mid-size teams. Ex-NASA engineers, senior-led, fixed price. Book a discovery call https://concepttocloud.com/contact ai aireadiness machinelearning dataengineering airoi https://video-meta.open.video/poster/TJuKodbO3ld6/AlhZP-VPik2_shRnLS.jpg https://streaming.open.video/contents/TJuKodbO3ld6/1790815545/index.m3u8 https://videos.concepttocloud.com/embed?contentId=AlhZP-VPik2 609 https://videos.concepttocloud.com/v/why-your-ai-is-stuck-in-a-proof-of-concept-and-how-to-ship-it Why Your AI Is Stuck in a Proof-of-Concept (and How to Ship It) Your AI demo works. Its been nearly ready for six months. The gap between a demo and production isnt the model its everything around it. Heres what actually has to change. See where yours is blocked: https://concepttocloud.com/services/ai-data-preparation/ai-readiness-audit Why most AI never ships, and how to get yours to production: - The POC graveyard: why demos stall - The real blockers data readiness, guardrails, security, evals, ownership - Our thesis: AI ships when the guardrails sit in the data model, not the prompt - The readiness checklist to get from demo to production Chapters 0:00 The demo thats been nearly ready for 6 months 1:15 Why POCs stall 95% became 80% 2:02 The real blocker: data that isnt production-ready 2:52 Guardrails, security and compliance 4:45 Evaluations how do you know the answers right? 5:49 Ownership: who runs it once its live 6:41 The readiness checklist 8:39 What happens when the LLM goes offline 8:55 Run the AI-Readiness Audit Concept to Cloud ships production AI in regulated, data-heavy industries guardrails in the data model, not the prompt. Run the free AI-Readiness Audit: https://concepttocloud.com/services/ai-data-preparation/ai-readiness-audit AI machinelearning production dataengineering MLOps https://video-meta.open.video/poster/QGjestH42paA/ABgYiFVyPk2_lZlNyY.jpg https://streaming.open.video/contents/QGjestH42paA/1790815543/index.m3u8 https://videos.concepttocloud.com/embed?contentId=ABgYiFVyPk2 578 https://videos.concepttocloud.com/v/define-once-use-everywhere-from-excel-to-product-ep-7 Define Once, Use Everywhere From Excel to Product (Ep. 7) The semantic layer is the point. In episode 7 we get experimental: multiple interfaces on the same data, so your whole organisation reuses one consistent, secure definition instead of re-inventing metrics. In this episode: Why reuse and consistency beat raw DB access Securing data behind a login and a model Guardrails that stop LLMs going rogue on your DB Chapters: 0:00 Why reuse your data 2:36 Secure access by design 5:06 Guardrails for LLMs Concept to Cloud: https://concepttocloud.com Saiku (open source): https://github.com/spiculedata/saiku dataengineering analytics excel saiku dataproducts BI https://video-meta.open.video/poster/TZiyFh0Gtdlr/AlMJiFozPA2_YCsAgd.jpg https://streaming.open.video/contents/TZiyFh0Gtdlr/1790815540/index.m3u8 https://videos.concepttocloud.com/embed?contentId=AlMJiFozPA2 488 https://videos.concepttocloud.com/v/rag-done-safely-chat-with-your-database-without-leaking-it RAG, Done Safely: Chat With Your Database Without Leaking It Row-level security stops the wrong ROWS from leaking. It says nothing about the wrong FIELDS. This build-along closes that gap: Tom takes a RAG chatbot that can already answer questions about a shared notes wall, locks it down so each person only ever retrieves rows theyre allowed to see then goes further and adds PII redaction so sensitive fields never reach the model in the first place, with Claude Code writing the fix live. Youll see: why the model only retrieves what youre allowed to see isnt the whole story, a real Postgres row-level-security policy added end to end, PII redaction added with one prompt and tested against real data, the trade-offs Tom flags honestly along the way (would I ship this? Absolutely not and why), and where this fits with the two earlier builds in the series. Chapters: 0:00 RAG, Done Safely: Chat With Your Database Without Leaking It 1:40 Row-Level Access, Then PII Redaction 2:56 Locking It Down to Row-Level Access 4:05 Dont Let It Walk Out the Back Door 10:48 I Have 2 Notes. The Wall Has 34. 12:33 Adding PII Redaction With One Prompt 13:56 Testing PII Redaction End to End 19:41 Hiding It in the UI 22:12 The PII Redaction Reveal 24:11 Would I Ship This? Absolutely Not. 25:09 Where to Go Next Is your AI agent actually ready for real data and real production traffic? Concept To Clouds free AI-Readiness Audit tells you exactly where the gaps are book a Discovery call: https://concepttocloud.com/services/ai-data-preparation/ai-readiness-audit --- Concept To Cloud helps regulated and mid-size teams take AI agents from toy demo to production real data sources, guardrails, and a plan for what happens when things break. concepttocloud.com https://video-meta.open.video/poster/Oinagfb5xhX6/6lgdj_piPR2_LiAIWt.jpg https://streaming.open.video/contents/Oinagfb5xhX6/1790816075/index.m3u8 https://videos.concepttocloud.com/embed?contentId=6lgdj_piPR2 1547 https://videos.concepttocloud.com/v/build-an-ai-agent-that-works-on-your-real-data-not-a-toy-demo Build an AI Agent That Works on Your Real Data (Not a Toy Demo) Most AI agent demos stop at a toy chatbot. This build-along goes further: Tom builds an AI agent on top of a real Supabase-backed notes app then deliberately tries to break it, and shows exactly how guardrails (a deterministic gate, not another LLM call) stop it going rogue. Youll see: the agent get built live with Claude Code, a working chatbot answering real questions about real data, an attempt to jailbreak it into answering off-topic questions, the guardrails fix that shuts that down and the boring but essential production questions most demos skip: what happens when your model provider goes offline? Chapters: 0:00 Build an AI Agent on Your Real Data 2:42 From a Supabase Notes Wall to a Real Agent 4:00 Asking Claude Code to Build the Agent 6:10 A Proof of Concept, Not an Exfiltration Risk 9:36 Testing the Working Chatbot 11:34 Trying to Fool My Own Agent 12:20 Only Answer Questions About the Data 14:21 3 Ways to Stop an Agent Going Rogue 16:05 Watch It Refuse 18:09 What Happens When Your AI Provider Goes Down? 19:22 Where to Go Next Is your AI agent actually ready for real data and real production traffic? Concept To Clouds free AI-Readiness Audit tells you exactly where the gaps are book a Discovery call: https://concepttocloud.com/services/ai-data-preparation/ai-readiness-audit --- Concept To Cloud helps regulated and mid-size teams take AI agents from toy demo to production real data sources, guardrails, and a plan for what happens when things break. concepttocloud.com https://video-meta.open.video/poster/VmDatf143d5r/A7MZj-VyP72_sASDbe.jpg https://streaming.open.video/contents/VmDatf143d5r/1790816058/index.m3u8 https://videos.concepttocloud.com/embed?contentId=A7MZj-VyP72 1214 https://videos.concepttocloud.com/v/excel-that-stays-fresh-from-excel-to-product-ep-9 Excel That Stays Fresh From Excel to Product (Ep. 9) Spreadsheets go stale this one doesnt. In episode 9 we tour the open-source Saiku repo, build a live query, and connect Excel to a live data source so the numbers are always current. In this episode: The open-source Saiku repo (and how to run it) Building a live query against the cube Excel connected to a fresh, governed data source Chapters: 0:00 The open-source Saiku repo 7:47 Building the query live 15:20 The problem were solving 20:27 Querying Saiku from Excel Concept to Cloud: https://concepttocloud.com Saiku (open source): https://github.com/spiculedata/saiku dataengineering analytics excel saiku dataproducts BI https://video-meta.open.video/poster/iibGodaPwpqB/6lhIiEVjPR2_VPUxcf.jpg https://streaming.open.video/contents/iibGodaPwpqB/1790816063/index.m3u8 https://videos.concepttocloud.com/embed?contentId=6lhIiEVjPR2 1354 https://videos.concepttocloud.com/v/how-technical-due-diligence-actually-works-from-someone-who-runs-them How Technical Due Diligence Actually Works (From Someone Who Runs Them) Most technical due-diligence reports are 60 pages nobody reads - and they miss the one thing that actually kills the deal. Heres what a technical DD really is, what it should find, and how to spot a checkbox exercise. Get a 10-minute read on your deals tech risk: https://concepttocloud.com/services/technical-due-diligence/tech-risk-scorecard Ive run technical due diligence on both sides of the table. In this video I break down what a real DD looks like for a private-equity deal team: - What a technical DD actually answers (hint: its not a code review) - The risks a good one surfaces - and the ones cheap ones miss - How key-person risk, scalability ceilings and AI theatre change the price - Good DD vs a generic template - including red flags in the DD provider itself - What it costs and how long it takes Chapters 0:00 The report nobody reads 0:22 Who this is for 1:00 What a technical DD actually is 1:57 What a good report finds 2:05 Key-person risk 2:48 Scalability ceilings 4:11 Security & compliance 5:51 Tech debt with a number 6:53 AI theatre 7:48 Integration landmines 8:53 Good DD vs a checkbox exercise 10:01 Cost, timeline, and what you get 10:35 Where to start Free: the Tech-Risk Scorecard: https://concepttocloud.com/services/technical-due-diligence/tech-risk-scorecard Pre-acquisition diligence checklist: https://concepttocloud.com/resources/pre-acquisition-tech-diligence-checklist Concept to Cloud runs board-ready technical due diligence for private equity - ex-NASA engineers, senior-led, fixed price. Book a discovery call: https://concepttocloud.com/contact privateequity duediligence techdiligence M&A privateequityinvesting https://video-meta.open.video/poster/Q_a8hfWGYxyX/7RhYiEVijR2_zVWDdc.jpg https://streaming.open.video/contents/Q_a8hfWGYxyX/1790816112/index.m3u8 https://videos.concepttocloud.com/embed?contentId=7RhYiEVijR2 668 https://videos.concepttocloud.com/v/one-model-four-apps-from-excel-to-product-ep-10 One Model, Four Apps From Excel to Product (Ep. 10) The finale: from spreadsheet to shipped product. In episode 10 Claude builds a cross-platform app on top of the same model we started with turning a static Excel file into something tailored, live and deployable. In this episode: Turning the Excel model into a real application Scaffolding a cross-platform app with Claude Deploying it and wrapping the series Chapters: 0:00 From Excel model to real app 5:08 Scaffolding the app 10:13 Deploying it 12:43 Series wrap-up Concept to Cloud: https://concepttocloud.com Saiku (open source): https://github.com/spiculedata/saiku dataengineering analytics excel saiku dataproducts BI https://video-meta.open.video/poster/i9ySAxvH_pcz/7lNZOEUjPR2_sWFvCj.jpg https://streaming.open.video/contents/i9ySAxvH_pcz/1790816305/index.m3u8 https://videos.concepttocloud.com/embed?contentId=7lNZOEUjPR2 938