Building products people can open and use — from shipped browser games to AI systems that reason, retrieve, and remember.
🎓 B.S./M.S. Computer Science, B.S. Electrical Engineering, B.S. Innovative Leadership & Engineering Entrepreneurship
University of Illinois Urbana-Champaign (Grainger) · Class of 2029
📍 Chicago, IL
I'm a CS / EE / ILEE student at UIUC focused on shipping software and AI products. I've worked at Databricks and Rivian. My flagship build is CoolGames.io — a free classic browser-games product — alongside AI work like Chronos and TeXForm.
- Agentic AI & reliable tool-using systems
- Retrieval, long-term memory, and context management
- LLM evaluation, grounding, and product reliability
- AI infrastructure & platforms
- AI product engineering (scope → ship → iterate)
Present · Focused on open-source AI / Omnigent
- Software Engineer Fellow on open-source AI agent infrastructure around Databricks Omnigent — a meta-harness for composing, controlling, and sharing agents
- Work spans agent composition across harnesses and models, policy/control at the meta-harness layer, and infrastructure that helps agentic products ship reliably
- Omnigent: github.com/omnigent-ai/omnigent · omnigent.ai
Present · Focused on Vehicle Telematics
- Software Engineer Intern on vehicle telematics — software that connects vehicle systems, data, and product surfaces
- Building in a high-reliability automotive product environment across systems and application layers
Free classic browser games — Chess, Tetris, Snake, Pong, 2048, Wordle, Ludo, and more. No downloads. No accounts. Just play.
Highlights
- 14 playable games across board, puzzle, and arcade
- Most titles support 2 Player or vs AI; Ludo is 4-player
- Per-game engines (HTML5 Canvas / DOM) inside a shared React + Vite shell
- Live on GitHub Pages
Play: a2105z.github.io/CoolGames.io · GitHub: a2105z/CoolGames.io
Products oriented around the same loop: reason → retrieve → remember (and act).
Natural language in. Constraint-backed calendar out. Chronos interprets planning intent with an AI/deterministic planner, then a verified scheduling engine places blocks — so the model never invents capacity the calendar can't support.
Highlights
- AI planner separate from the SchedulingEngine source of truth
- Multi-spot NL (“find me 3 spots, each 2 hours”); FullCalendar; JWT multi-user; America/Chicago time rules
- Stack: Python · FastAPI · React · TypeScript · Docker · CI
GitHub: a2105z/Chronos
Handwritten notes → LaTeX. Vision/OCR pipeline productized as React + FastAPI with a live Hugging Face demo.
Live demo: Hugging Face Space · GitHub: a2105z/TeXForm
| Project | What it is |
|---|---|
| Meridian | Full-stack academic journey tracker — awards, scores, goals, auth-scoped data, analytics, export (FastAPI + React) |
| Cosmos | TI-84–inspired browser calculator — calc, graph, and AP Stats modes (live) |
Software Engineering, AI Engineering, and AI Product internships, especially:
- Agentic AI & AI infrastructure
- Retrieval / memory / context systems
- LLM platforms and evaluation
- AI product engineering
If you're building in these areas, I'd love to connect.