Systems

Systems, told through their decisions.

No feature tours. Each case study shows the constraints, the decisions with the alternatives I didn't take, the trade-offs, what failed, and what I learned from it.

Case studies

grex
orchestration · agents · desktop

One workspace to run Claude, Codex, Cursor and Gemini agents in parallel — and keep their work from colliding.

Case study · repo ↗
GitHub ↗
Jev RAG Benchmark
evaluation · retrieval · LLM systems

A free English benchmark that asks whether TypeSafe's Jev 1.13 actually improves a RAG pipeline — frozen candidate pools, calibration, paired confidence intervals, and the runs where the answer was no.

Case study · repo ↗
GitHub ↗

In the queue

meetwit A meeting assistant that transcribes on your machine and answers from your documents — nothing leaves the device. local-first · speech · retrieval
inclave Private AI over your local files: PDFs, spreadsheets and code, reasoned over entirely on-device with Ollama. local-first · inference
claude-remind-mcp An MCP server that indexes your Claude Code history with local BM25 so any past decision is one search away. tooling · retrieval · MCP
BioAgents A multi-agent research framework that forms hypotheses, reads the literature and tests them in the life sciences. multi-agent · research

grex

Jev RAG Benchmark