Computer Science at VIT Vellore. I build machine-learning systems end to end and like measuring them honestly.
Portfolio: adityasridhar.vercel.app: nine neural networks trained from scratch that run in your browser, plus interactive demos of the projects below.
- 🦯 BlindAssist: offline camera assistant for blind users; two YOLO detectors and a decision layer that says one thing at the right moment. Try it
- 🏢 Amazon ML Challenge 2026: rank 696 of 89,000+, entity resolution over 10M business records
- 🧬 Protein impact explorer: Best Project, Code for Life hackathon (VIT)
- 🎙️ Offline voice assistant: speech in, speech out, first word in ~1.3 s, nothing leaves the laptop
- 🌐 The portfolio itself: Next.js, ONNX Runtime Web, every metric traced to its training log
Research. First-author paper at SIPCOV-2026 (solar-aware load forecasting), a second under revision at ICMLDE, two published Indian patent applications and more filed.
Open source. Merged fixes in Kornia, Hugging Face timm and deepset Haystack.
Tools. Python · PyTorch · ONNX · scikit-learn · LightGBM · FastAPI · TypeScript · Next.js · Docker · AWS
