Project video reel showcasing Terminal Notes, Status Panel, and Portfolio Kit.
Neo Coworker is a file-system-based agent runtime and desktop app I built for real day-to-day work. It supports tool-use, agent skills, permission, subagents, context management, and folder-scoped workspaces; it also uses execution traces and eval artifacts to capture model requests, tool calls, permission waits, retries, and terminal paths so long-horizon agent tasks can be reproduced, debugged, and regression-tested.
Visit projectAn automated RAG chunking and pipeline optimization experiment: from benchmark construction, document loading, chunking / embeddings, retrieval, hybrid search, reranking, to end-to-end metric evaluation. It lets an LLM agent adjust chunk size, overlap, top_k, rerank, prompts, and temperature inside a constrained mutation surface, then compares configuration gains through snapshot / rollback and telemetry.
Visit projectAn LLM-powered generative interactive fiction runtime.
Visit project
