Applied AI Engineer · Solutions Architect · Open Source Builder
I turn ambiguous customer problems into production AI. From requirements through architecture, implementation, and iteration. LLMs, computer vision, recommender systems, edge deployment. 3x hackathon winner (MIT, Harvard, UMass). Claude Code power user shipping open-source developer tools and agent skills.
- Applied AI Engineering → LLM post-training, fine-tuning, eval pipelines, agent orchestration, RLHF, RAG systems
- Solutions Architecture → customer requirements to shipped systems, end-to-end design across backend, inference, frontend, and edge (WebGPU, ONNX)
- Developer Tooling → open-source skills, agent frameworks, Claude Code ecosystem
Built Catalyze, 1st Place at the MIT LLM Hackathon:
- Multi-agent chemistry assistant with ChEMBL integration
- Lab automation workflows and safety analysis
- Full system built end-to-end in under 36 hours
- The Paddock - F1 encyclopedia with 860+ drivers, race replay, telemetry compare, and an ONNX-powered what-if championship simulator. Live
- alignrl - LLM post-training library: SFT, GRPO, DPO, eval, and inference in one
pip install. PyPI · Docs - SkillKit - Collection of AI coding skills for Claude Code, Cursor, Windsurf, and Copilot. Gallery
- Onyx - Google Gemma 4 running fully in the browser via WebGPU. Multimodal chat and E2B vs E4B side-by-side arena. Live
- ResearchProf - Wikipedia-style professor discovery powered by OpenAlex (240M+ works). Live
- Claude Code Hub - GitHub-style activity tracker for Claude Code with session timelines and contribution graphs
- today-i-learned - A running log of things I built, broke, and figured out. Diagrams, annotated scripts, runnable demos
Hackathons → 1st Place MIT LLM Hackathon (Catalyze) · Harvard Hackathon winner · UMass Hackathon winner
Open Source → alignrl (published on PyPI) · SkillKit · contributions to llama.cpp, torch-sim, and the Claude Code skills ecosystem
💼 Open to Applied AI / ML Engineer / Research Engineer roles