CS (ML & Data Science) at the University of Indianapolis. I build local-first AI tools and report where each one fails.
Currently: LLM evaluation and benchmark design, Handshake AI (Mar 2026 to present).
- same: find every photo that contains a specific object, on-device. DINOv2 candidates, SIFT/RANSAC verification.
- llm-ladder: runs a prompt on a small local LLM first and moves to a bigger one when repeated answers disagree.
- Cutroom: desktop video editor in Rust (Tauri) with local AI models. Installers for macOS, Windows, Linux.
- agent-office: my fork of a 3D workspace for coding agents. Added a low-power graphics mode, a city, driving physics, a race track and an arena.
- VisaRadar: checks whether an employer sponsors visas, using real Department of Labor filings.
- homeground: type your address, ride a longboard down your own street. OpenStreetMap data rebuilt as a 3D level.
Python · PyTorch · TypeScript · Rust · FastAPI · Three.js / WebGPU · Ollama · scikit-learn
Indianapolis, IN · open to Summer 2027 internships · LinkedIn · psingh@uindy.edu