BM25 beat my vector database (sometimes)
A crossover framework for lexical versus vector retrieval on code — and the adversarial bench harness I built so my own argument can lose.
sam@latino:~/writing$ ls -lt
writing
Notes from building agent systems, evals, and self-hosted LLM infrastructure. Every post is backed by a repo you can clone and a case study you can check it against.
3 posts rss
Every model fails tool calling differently
Tool calling is the load-bearing primitive of every agent stack, and open models break it in at least eleven distinguishable ways. Naming the failure modes changes how you build the layer above.
Red-teaming my own agents with the OWASP Agentic Top 10
Turning "resists prompt injection" into a regression number: a deterministic harness, 146 probes across five OWASP agentic categories, and a hardening sweep that went 73% → 3% → 0%.