fastpaca
Practical engineering and thoughtful research on the problems that sit between people and technology.
Research · Open Source · Fractional & Advisory
Benchmarking LLM agents shouldn't mean wrestling with brittle scripts and lost progress. pacabench is a local-first tool that provides reproducible, reliable benchmarks with isolated execution, persistent state, and built-in metrics tracking—no SDK required.
I benchmarked Mem0 and Zep on MemBench to understand why production agents were failing. Memory systems cost 14-77x more and were 31-33% less accurate than naive long-context.
Read more →An introductory guide to how LLMs handle 'memory', from context windows to retrieval systems and everything in between.
Read more →Why we built infrastructure for LLM context management, and what problems it solves.
Read more →I work with companies at critical technical inflection points.
If you're facing a hard problem at the intersection of technology and organisation, let's talk.
Users expect full message history, but LLMs have hard limits. Most teams rebuild the same Redis + Postgres stack; context-store extracts that pattern into a reliable Elixir service with Raft consensus, horizontal scaling, and deterministic compaction.