I build things that think — from neural nets written by hand to AI tools that solve real-world problems.
Smart India Hackathon 2025 | AI-Driven Fleet Induction Planning
Automates the nightly scheduling of 25 Kochi Metro trainsets using multi-objective optimization (NSGA-II) and Explainable AI (SHAP/LIME). Manages 6 interdependent variables — fitness, maintenance, branding, mileage, cleaning, and stabling — to generate transparent, optimized decisions in real time.
Python · pymoo · scikit-learn · Firebase · SHAP/LIME · Flask
A production-ready Model Context Protocol server for AI-powered document intelligence
Plugs into AI platforms via the MCP protocol to provide smart document processing: summarization, key point extraction, context-aware search, and multi-format support (PDF, DOCX, TXT, ODT and more). Session-managed, token-authenticated, and built for real-world AI integration.
Python · FastAPI/ASGI · MCP Protocol · Puch AI · Vercel
Deep learning, demystified — no frameworks, just NumPy
Built a full multi-layer neural network from the ground up — perceptron to backpropagation — with zero black boxes. Trained and evaluated on Iris and MNIST datasets, complete with loss curves, confusion matrices, and prediction visualizations.
Python · NumPy · Jupyter Notebook · scikit-learn · matplotlib
Low-level systems programming from an ICT internship
Hands-on C++ projects exploring systems programming concepts, data structures, and algorithms as part of a structured ICT internship.
C++
| Area | Focus |
|---|---|
| 🤖 AI / ML | Building models from scratch, LLM fine-tuning, optimization algorithms |
| 🛠️ AI Tooling | MCP servers, AI-integrated backends, document intelligence |
| 🏆 Hackathons | Smart India Hackathon, rapid prototyping, real-world problem-solving |
| 💻 Software Dev | Python, C++, Flutter, Java, full-stack applications |
| ☁️ Cloud & Data | Firebase, scalable architectures, data pipelines |
Languages:
AI / ML:
Frameworks & Tools:
"The only way to do great work is to love what you do." — Steve Jobs
Explore my repos, open an issue, or just say hi — always happy to collaborate! 🚀