Open source
Useful pieces left on the workbench.
Agents, RAG systems, developer tooling, and small explorations published so someone else can start further ahead.
Repositories (5)
Tokenius
A coding agent built from scratch in the terminal — no frameworks, just TypeScript and LLM calls — to understand how tools like Claude Code actually work under the hood.
Agentic RAG
Production-grade RAG that goes beyond naive retrieval — hybrid search, LLM reranking, query reformulation, and explicit failure handling.
Bunbase
Strict TypeScript starter optimized for AI-assisted development — five parallel checks on every commit so agents get instant feedback.
Tech Radar Agents
Five AI agents collaborate to evaluate technology adoption decisions — no frameworks, just TypeScript, OpenAI SDK, and ~500 lines of code.
MCP Prompt Lab
Prompt evaluation workbench as an MCP server — run evals, compare providers, and test assertions directly from Claude Code.