AI Product Manager and hands-on LLM engineer — RAG, agent orchestration, and the Python behind them
LLM engineering across retrieval-augmented generation with vector and hybrid search, agent and retrieval orchestration on LangChain and LangGraph, Model Context Protocol tooling, LangSmith observability, and evaluation-driven delivery of production AI systems.
MCP end to end — servers published to PyPI and the official registry, clients wired into agent runtimes across four transports, an authenticated gateway in front of internal servers, and protocol-level defects reported upstream, one of them already fixed and released.
PythonProduction Python for AI systems — async FastAPI and FastMCP services, LangChain and LangGraph retrieval pipelines, tested with pytest and shipped through automated release and deployment pipelines.
Claude CodeProfessional-grade command of Claude Code — orchestrators, subagents, dynamic workflows, hooks, skills, slash commands, MCP, and context engineering — applied to production multi-agent development environments.
Expertise in developing and launching AI-powered products, combining strategic vision with Agile practices and robust market and user analytics.
Relational database design and SQL, plus hands-on work with the SQLite, PostgreSQL, MongoDB, and Chroma stores behind AI retrieval systems I build — pgvector and sqlite-vec, full-text, semantic, and hybrid search.
7,508 contributions last 12 months
An opinionated, hardened Claude Code environment that extends the harness rather than replacing it: curated skills, always-on rules, quality-enforcing hooks, MCP servers, durable cross-session memory, and a quality toolchain, with deliberate model, effort, and permission defaults. It rides on Claude Code's native dynamic-workflow orchestration and supplies the reliability and standards layer around it. Installable via Claude Code Toolbox, backed by MCP Context Server, with quality gates enforced at runtime.
A high-performance Model Context Protocol server that gives LLM agents a durable, searchable memory layer beyond the context window. Thread-scoped storage over SQLite or PostgreSQL, with semantic, full-text, and hybrid search, cross-encoder reranking, pluggable embedding and summary providers, metadata filtering, and batch operations — all behind a standard MCP interface.
A cross-platform installer and declarative environment-configuration framework for Claude Code. One YAML file captures all agents, MCP servers, slash commands, hooks, skills, rules, system prompts, and settings; one command installs everything on Windows, macOS, or Linux — portable, reproducible, centrally updatable.
A research and educational quality-assurance tool that gives a chest X-ray draft impression a transparent second read. Two tiny 4B models — a fine-tuned MedGemma that grounds the image into labeled findings with bounding boxes and an NVIDIA Nemotron-3 Nano 4B that parses the draft into the same labels — feed a deterministic, model-free comparator that flags MISSING, UNSUPPORTED, and URGENT discrepancies, with every flag traced to a specific finding, a specific phrase, and a box on the image. Not a medical device and not a diagnostic tool.
Certificate of participation in Build Small, a Hugging Face × Gradio hackathon celebrating small, local, open-weight models — building real, tinkerable apps powered entirely by models under 32B parameters.
The certification validates my knowledge of the Scrum framework, the Scrum Master accountabilities and how to apply Scrum.
A validation of acquiring in-depth knowledge and practical skills in managing software projects, focusing on Agile and Scrum methodologies, risk assessment, team collaboration, and effective communication.
A comprehensive certification validating expertise in Agile project management principles and proficiency in Atlassian tools for effective project execution.
Every page on your site is two documents: the one a browser renders for people and the one a parser reads. AI agents read the second one. Here is what I learned making this site legible to them -- and what a machine reader finds on the personal sites of eight of the best-known people in AI, from a hidden Order of the Unicorn to a robots.txt that redirects into itself forever.
For a few weeks now, Claude Code has been writing its own multi-agent harness for each task -- and it quietly retired the orchestration code I used to maintain by hand. Here is what changed, the patterns worth knowing, and the research idea I think sits underneath it.
A Field Notes write-up on building a chest X-ray draft auditor from open data alone: two tiny 4B models, a deterministic comparator, and the evaluation-integrity lesson that nearly cost me the better model.
Your AI coding agent loses its plan every time the context window compacts. MCP Context Server fixes that with persistent, searchable memory -- one Docker command and your agents remember everything.