Machine learning engineer and full-stack developer · B.Tech in Computer Science and Engineering at NIT Patna (2024–2028)
- Freelance ML engineer, shipd.ai (Project Eris): solved 100+ research-grade ML challenges end to end across computer vision, NLP, speech, signals, graphs and scientific data, shipping 90+ reproducible pipelines graded on hidden test sets.
- Freelance AI benchmark engineer, shipd.ai (Quest Olympus): author long-horizon coding tasks on open-source Python libraries (dask, sympy, great-tables, trimesh, networkx) that evaluate frontier AI coding agents.
- Smart India Hackathon 2025: national finalist.
| Project | What it does | Result |
|---|---|---|
| methane-plume-footprints | Outlines confirmed methane plumes in remote-sensing scenes and rejects false alarms (PyTorch, CNN trained from scratch) | 0.68 graded vs 0.58 for the reference network |
| cross-script-thread-recovery | Untangles interleaved Wikipedia talk-page conversations; trains on Greek, scored on Chinese (LightGBM, learning to rank) | 0.54 CV vs 0.32 published reference |
| acg-ant-gate-crossings | Sub-pixel ant detection and tracking that logs every crossing of a virtual gate in video (PyTorch U-Net) | 1.83 px localisation |
| schematic-to-board-layout-retrieval | Matches a KiCad schematic to its PCB layout among 20 candidates (two-tower CNN, InfoNCE) | MRR 0.175 → 0.212 |
| remote-gpu-setup | Lends a Windows PC's GPU to a collaborator: WSL2 + CUDA PyTorch, SSH restricted to a private Tailscale network | Infrastructure |
| health-care | Full-stack healthcare web app with authentication and email verification (Node.js, Express, Docker) | Live demo |
- Machine learning: Python · PyTorch · scikit-learn · LightGBM · XGBoost · Hugging Face Transformers · OpenCV · NumPy · pandas · SciPy
- Web: Angular · React · Next.js · Node.js / Express · Laravel · TypeScript · MySQL · MongoDB · Firebase
- Infrastructure & tools: Git · Docker · Linux · SSH · Tailscale · WSL2 + CUDA · Kaggle · Blender · Three.js