What I Build
Intelligent systems, end to end
Agentic AI Systems
LangChain, RAG, multi-agent loops. Shipped on top of 200M-row production datasets, not demo notebooks.
Inference Infrastructure
vLLM, TGI, RouteLLM, Langfuse, Docker/BuildKit. Enterprise LLM serving built to hold up under real load.
Full-Stack AI Products
Next.js, FastAPI, DuckDB. I build the whole thing — frontend to data pipeline.
Published Researcher
First author, IEEE TENSYMP 2023. Co-author, Springer 2022. Deep RL research at Rutgers RUCI Lab.
Beyond the terminal
The things that shape how I build
Spending a Saturday rebuilding a Technic drivetrain and rewatching onboard lap footage are not separate hobbies from my work. They're the same brain.
Sector 1
Formula 1
I watch qualifying like other people watch sport. I actually look at lap delta charts. I have real opinions about floor design and have definitely lost sleep over DRS zones.
F1 is fundamentally about finding time — sector by sector, corner by corner, tenth by tenth. That thinking shows up in how I work. When something is slow, I want to know exactly where. Profiling, tracing, latency budgets — these aren't chores. They're the interesting part.
Sector 2
LEGO
Technic and Creator sets. The structural problem-solving is genuinely fun — how do you build a working differential with 12 pieces? A Technic gearbox is a mechanical API. Each brick has a contract. Snap wrong ones together and the whole thing collapses.
A Dagster asset has a contract too. So does a LangChain tool. I care about those boundaries for the same reason I care about brick compatibility — get them wrong and nothing works right. I've built F1 cars in LEGO. I'm working on Spa-Francorchamps in Minecraft. Eau Rouge is harder than it looks.