DAIS 2025 is six weeks out. I've been collecting notes on what separates the enterprise AI teams that are actually shipping production systems from the ones that are still stuck in POC purgatory. Here's the checklist I use before any AI initiative gets to production — not
If you want a model to behave consistently across a long engagement — maintaining a specific perspective, applying a specific methodology, staying within a defined scope — you give it a persona. That's the idea, anyway. The execution turns out to be more interesting than the idea. I started experimenting
It's April 2025, two months before DAIS 2025, and I've been running a list of what Databricks hasn't solved from the 2024 announcements. This isn't a complaint — the platform has gotten better every quarter. But honest critique is how you calibrate what
AI security became a board-level conversation in 2024. Not because AI security is new — adversarial ML attacks have been documented for years — but because AI systems are now making decisions that matter: credit approvals, insurance pricing, fraud flags, hiring screens. The attack surface has expanded to the point where
The goal I had been working toward since month four was simple to state and hard to execute: make the AI tools I use produce better output without requiring me to write better prompts. The knowledge retrieval system was the storage and retrieval half of that goal. Automatic prompt refinement
The governance conversation in data engineering has historically centered on data: who can access what table, what's the lineage of this column, does this dataset comply with GDPR. That conversation is mature — the tooling is good, the processes are established, most serious data organizations have solved it or
Databricks rebranded themselves as a "Data Intelligence Platform" at DAIS 2024, and they've been using the term consistently since. It's a more ambitious claim than "data lakehouse platform" — and it invites examination. What does "data intelligence" actually mean, and does
When you read about pgvector, you get the idealized version: add the extension, store vectors, query by cosine similarity, done. Production use adds the parts the tutorials leave out — index tuning, embedding model choice, the hybrid search problem, and the edge cases that only appear when your knowledge base has
Six months into working with UniForm and the broader interoperability story from DAIS 2024, and I want to document what's actually happening in production. The interoperability revolution that was announced at Summit is real — it's also slower and more complicated than conference demos suggest. What'
DAIS 2024 had a lot of sessions on training programs, certifications, and emerging job functions. The subtext underneath most of them was the same: the skills floor for data practitioners is rising, and it's rising fast. If you're not building that floor under yourself right now,
The noise around DeepSeek R1 was hard to separate from the signal for about 48 hours. Once the dust settled, the actual result was clear: a Chinese lab trained a reasoning model that matches OpenAI o1 on hard benchmarks, and they did it at a fraction of the reported cost.
The prototype worked. That meant it was time to stop prototyping and build the real thing. What had started as a PostgreSQL database with a JSONB blob and a pgvector column needed to become something I could rely on across multiple concurrent client projects, maintain without weekly intervention, and extend