Seven years with 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. I wrote down what I think drives that.
Semsudin Field Notes
Notes from building
production AI.
Essays on the architectures, patterns, and tradeoffs I run into on enterprise GenAI projects.
RSS feedAll field notes
-
Seven years with 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. I wrote down what I think drives that.
-
The cost of self-healing QA (real numbers)
I ran AI-driven self-healing on 1000+ tests for a global pet care company. This is what it cost and what it saved.
-
Shipping GenAI that survives production
I architected a multi-agent platform at a global industrial company. This post covers why most enterprise AI never reaches production and what the projects that do reach it have in common.
-
Image promotion: the prod isolation pattern I use for enterprise AI
Many CI/CD pipelines let production pull whatever the build server last produced. That is fine for a marketing site, but on an enterprise AI platform it leads to 2am rollbacks. This is the image promotion pattern I use on those platforms.
-
AI agents in the QA loop
This post shows what it looks like when AI agents are integrated into the development cycle for test generation, self-healing tests and PR fixes. In most 'AI for QA' tools, the AI is bolted on.
-
Why we picked pgvector over Pinecone
Enterprise RAG projects usually start by asking whether they need a dedicated vector database. On the enterprise GenAI platform I worked on, we chose pgvector. This is the decision framework I used.
-
The RAG patterns I keep reaching for
This is the decision tree I use when designing retrieval for enterprise AI. I start with vanilla pgvector and only add hybrid search or knowledge-graph traversal when the use case needs it.