01
Document comparison at scale
10-user proof-of-concept to a production system serving 350 concurrent users on 50 MB files — event-driven, fully observable.
Tech Lead, Data & AI — AXA XL
Denis Baciu
I industrialise AI in financial services — taking proof-of-concepts and turning them into production systems people actually rely on.
Proof of conceptthe demo that works
AI Lab pilotobservability, audit, edge cases
Production systempeople actually rely on it
01 / Latest writing
full archive →- 2026-08 MCP session handles are capability tokens, not pointers linkedin
- 2026-08 Why Regulated Enterprises Will Mix Open and Closed Models linkedin
- 2026-08 Multi-cloud AI observability needs a common transponder linkedin
- 2026-08 AI-assisted PostgreSQL migrations need a human review loop linkedin
- 2026-08 The Real Bottleneck in Data Platforms Is Neglected Orchestration linkedin
02 / Projects
Shipped work, described at the level already public on my CV. Systems, not demos.
02
Hybrid semantic search
Org-wide ingestion into vector search, upgraded to hybrid semantic + keyword — underwriters find the right document in seconds.
03
AI code review
Internal PR-review agent wired into GitHub and CI/CD, flagging issues before human review — cutting review time, lifting quality.
03 / Now
- Building the production-assurance layer for AI systems at work — evals, tracing, cost and rollback for agentic workflows in a regulated environment.
- Preparing one conference talk on making AI agents pass production standards in financial services.
- Consolidating everything I've ever published into this archive, so it stops depending on anyone else's platform.
- Protecting the fixed slots: training, swimming, family evenings. Side work fits the residual, never the reverse.
04 / About
I started in Chișinău — applied informatics, then a master's in software — writing C#/.NET for logistics systems. The route since ran through Belgium, the Isle of Man, Birmingham, and the London market: pensions platforms, insurance comparison, and now specialty insurance at AXA XL, where I lead Data & AI engineering and an AI Lab of 15+ engineers and data scientists.
The pattern across fourteen years is the same one twice a decade: a promising system meets messy reality. That last mile — observability, audit, actual users — is my specialty, and in financial services it's the mile that matters.