CS, Mathematics, and Data Science triple major at UW-Madison, expected May 2028. I build AI systems end to end: retrieval over messy data, multi-agent pipelines, and the GPU kernels underneath them.
AI Engineer Intern at RIA Advisory. Building an agentic system that turns unstructured client input into cited business requirements documents. It runs GraphRAG over Qdrant and Neo4j behind a LangGraph multi-agent pipeline.
CDIS & OpenAI Summer AI Lab. Writing fused CUDA kernels for linear attention, targeting roughly 15x longer usable context on an H100.
- race-attention-kernel - why an ICLR 2026 linear-attention repo's shipped CUDA kernels never run, with A100 baseline benchmarks; the replacement fused kernel is in progress.
- Tripwire - prompt-injection security testing for AI agents, published on PyPI as
tripwire-eval, so you can find out whether your agent is hijackable before it ships. - Wattbot - hybrid RAG system built for a Kaggle competition, reaching 87% accuracy across 282 test queries.
- AI_Interviewer - mock interview agent, 3rd place at MadData 2025.
- CustomShell - a POSIX shell written in C: pipelines, redirection, job control, and builtins.
A lot of what I spend my time on is not on this profile. The RIA Advisory system above is private company code. So is my work at the UW Tech Exploration Lab on a voice agent and its dashboard, and my work at the UW Data Science Institute building social network analysis tooling for the Wallace Foundation initiative. Those projects sit in private or employer-owned repositories, so they are described on my resume and LinkedIn rather than here. Happy to talk through any of them directly.
- Email: agrawal54@wisc.edu
- LinkedIn: https://www.linkedin.com/in/arunjay-agrawal-aab235322