Writing / 06
Technical decisions, experiments, lessons, and ideas that are still taking shape.
Most agent frameworks have no way to tell if your agent is getting better or worse. I built an evaluation system into Pori that tests accuracy, reliability, performance, and safety. Here's why each matters and how they work.
How I built AuditRAG, a fullstack Retrieval-Augmented Generation platform with hybrid search, auto-evaluation, and per-query observability. The architecture decisions, tradeoffs, and what I learned.
How we replaced LangChain with ~200 lines of protocol-based LLM wrappers in Pori. The architecture, the tradeoffs, and why direct SDK integration wins for agent frameworks.
The software 'app' container is dissolving. Capabilities will soon live everywhere, summoned on demand.
Here is the clean and efficient workflow I use When working on OSS,that helps me in sync with the project while keeping your contributions organized.
My first post on my new blog.