Data science and ML student from Jaipur, India 🇮🇳. I build end-to-end ML projects, from the model to a deployed app, and contribute to open-source Python libraries.
- 🧠 Built a neural network from scratch in NumPy (hand-written backpropagation, Adam, calibration) that predicts customer churn and how much a retention offer helps
- 📊 Interested in causal inference, explainable ML and LLM tooling
- 🌍 11 merged pull requests in open-source projects including kornia, aeon, scikit-hep/uproot5 and FOSSASIA
Live demo · Code · How the model works
Predicts which telecom customers will leave, explains why, and decides who is actually worth a discount using causal uplift modelling.
- ChurnUpliftNet, my own model written from scratch in NumPy: one network predicts churn risk and the offer's effect. It matches XGBoost on churn (ROC-AUC 0.846 vs 0.847), verified by a gradient-check test
- Uplift targeting earns 2.4x the value of risk-based targeting at the same budget (simulated experiment)
- FastAPI backend, custom light/dark dashboard, Docker, GitHub Actions CI, 24 tests
| Project | What it does | Stack |
|---|---|---|
| exam-prep-ai | RAG over handwritten notes with page citations, AI-graded mock exams, spaced-repetition flashcards, PYQ-based study plans | Python, Gemini, FAISS, OCR |
| ai-trading-platform | TradePulse: AI trading dashboard for Indian stocks with live NSE prices, technical signals and news sentiment | Python, HTML/JS |
| focusly-app | Mobile-friendly to-do list and meal tracker with a nutrition calculator and offline Android install | HTML/JS, PWA |
11 merged pull requests across 7 projects:
