§ Blog
Field Notes
Notes from the edge
of production AI.
Opinionated essays on the architectures, patterns, and tradeoffs I encounter building enterprise GenAI in the field.
007
April 8, 2026
6 min
Seven years, one client: what repeats me
The same client has kept hiring me across four different projects, from a Selenium migration to a multi-agent GenAI platform. Here's what I think actually drives that, and why most consultants never see it.
006
March 24, 2026
7 min
The cost of self-healing QA (real numbers)
Everyone talks about AI-driven test automation. Almost nobody publishes the actual economics. Here's what self-healing tests cost and saved across 1000+ tests at Mars Petcare.
005
March 10, 2026
7 min
Shipping GenAI that survives production
Most enterprise AI never reaches production. The ones that do share a playbook. Here's what I learned architecting a multi-agent platform at Alfa Laval.
004
March 4, 2026
6 min
Image promotion, not direct builds: the prod isolation pattern enterprise AI needs
Most CI/CD pipelines let production pull whatever your build server last produced. That's fine for marketing sites. For enterprise AI platforms, it's a recipe for 2am rollbacks. Here's the pattern I use instead.
003
February 22, 2026
6 min
AI agents in QA: past the hype, into the loop
Most 'AI for QA' tooling is bolted on. Here's what it looks like when AI agents are integrated into the development cycle — from test generation to self-healing to PR fixes.
002
February 10, 2026
7 min
Why we picked pgvector over Pinecone
Every enterprise RAG project asks the same question: do we need a dedicated vector database? For Alva we chose pgvector. Here's the actual decision framework, not the marketing version.
001
January 18, 2026
8 min
The RAG patterns I keep reaching for
Not every RAG needs to be hybrid. Not every hybrid needs to be graph. Here's the decision tree I use when designing retrieval for enterprise AI — from vanilla pgvector to knowledge-graph traversal.