Where I build, break, and figure things out
The Lab is my personal AI engineering workspace — a collection of experiments, systems, and tools I'm actively building or have shipped. Everything here is real work, not demos.
Current focus
What I'm actively working on right now.
Multi-agent orchestration framework
Building a lightweight orchestration layer for coordinating specialised sub-agents — planner, researcher, writer, critic — with shared memory and structured handoffs.
RAG pipeline with hybrid retrieval
Combining dense vector search with BM25 keyword retrieval and a cross-encoder reranker. Testing on a private document corpus to benchmark retrieval quality.
MCP server for internal tooling
Exposing internal tools — calendar, task manager, knowledge base — to LLMs via the Model Context Protocol. Goal: a personal AI assistant with real context.
Explore the lab
How I work
Ship early, eval often
I'd rather have a working v0.1 with evals than a perfect v1 that ships in six months. Evaluation frameworks go in before features do.
Build to understand
Reading papers and watching lectures only sticks when I follow up with a working implementation. Every concept on this site has been through that filter.
Document the failures
The interesting stuff is usually what didn't work. I try to write up failures as honestly as successes — that's where the real learning is.
Follow the work
I write about what I'm building in the Engineering Journal — architecture decisions, experiments, and lessons from shipping AI systems.