Mihir Shrivas

I'm a software engineer and Data Science student who likes turning messy, half-formed ideas into systems that actually work — especially where ML meets real engineering. I care about building things that are reliable and transparent enough that users trust them and teams can move fast on top of them.

Most recently, I worked as a Tech & Data Project Assistant (contractor), where I audited a stealth startup's codebase and backend architecture, flagged security gaps, and built an AI-assisted hackathon judging MVP from scratch. My FastAPI-based "Hackeroos AI Agent Judge" ingests submissions via DevPost CSV, scores them with AI across weighted criteria, and ranks them on a live leaderboard — tested on 42 real submissions, with every score backed by stored evidence for full transparency. Before that, as an Associate Software Engineer, I built onboarding and market search APIs for a live trading app (React, TypeScript, Redux Toolkit), and shipped interface improvements that lifted customer engagement and retention by 15%.

I've also built applied ML products end-to-end — including a Dyslexia & Dysgraphia Detection Platform (Python, OCR, computer vision, Streamlit + ReactJS) that I tested directly in schools with educators and students, and published at DOSCI 2024. I'm drawn to problems where getting it right actually matters to the people using it not just the metrics.