- 🔭 Currently working on AI agents, GPU-accelerated physics simulations, Cargo attached to an Active Brownian Particle and RAG pipelines
- 🌱 Deepening my work in PyTorch internals, transformers, and ML systems design
- 🎓 M.Tech, AI for Sustainability, IIT Kanpur (2025-2027)
- 🎓 GATE 2025 AIR 210 (Data Science & AI)
- 👨🏫 Teaching Assistant for IDC612: Maths and Computation using Python
- 🧭 DPGC Student Representative, representing PG students at KSS
- 📫 Reach me at karing.iitk@gmail.com
- 📄 Resume
| Project | What it does | Stack |
|---|---|---|
| Decoder-Only Transformer | GPT-style character-level language model built from first principles — attention, causal masking, pre-norm residual blocks — with no high-level transformer libraries. 0.82M params, val loss 4.33 → 1.67. | PyTorch |
| MATLAB → Python Agent | Self-correcting agent that transpiles MATLAB to Python and iterates on its own test failures. Cut benchmark steps from 4 → 2 after a regex fix. | Python, LLM agent loop |
| Tool-Using Math & Search Agent | Deployed agent that routes queries to calculation and search tools. | FastAPI, Gemini API, Netlify |
| Turbulence Simulation (Sabra Shell Model) | GPU-accelerated solver using an integrating-factor RK4 scheme, with publication-quality analysis plots. | CuPy, CUDA, NumPy |
| Active Brownian Particle Simulations | GPU and pure-NumPy ABP simulations with FFT-based MSD analysis. | CuPy, NumPy, Matplotlib |
| Kedarnath Flood Reconstruction | ML pipeline reconstructing flood extent from satellite imagery. | Random Forest, GIS |
| AQI Prediction | Time-series imputation and forecasting on sparse sensor data. | BRITS, GAIN, XGBoost |
| Placement Portal | Full-stack portal serving 400+ users. | MERN |
