I'm a 3rd year B.Tech (CS, AI & ML) student at Atlas SkillTech University, Mumbai with a 9.5 CGPA and a co-authored research paper published at ICICC 2025.
I build things that sit at the intersection of real-world problems and AI - RAG pipelines on AWS, LLM agents, OCR-driven healthcare systems, and ML models trained on scientific datasets. I care less about following tutorials and more about shipping systems that actually work end-to-end.
Currently focused on: LLM-powered automation, AI agents, and accessible AI products for India.
"Build for the community."
AI / LLM
ML / Data
Backend / Infrastructure
Other Languages
OCR + Gemini + Random Forest - WHO-aligned health risk estimation for India
Built a system that reads uploaded blood reports via OCR, extracts structured medical values using Gemini 1.5 Flash, runs them through a trained Random Forest classifier, and returns explainable risk scores for diabetes, hypertension, and cardiovascular disease - all aligned to WHO South Asia thresholds.
Stack: Python · FastAPI · Gemini API · Tesseract OCR · Scikit-learn · Vite
Live: digital-twin-health.vercel.app | Repo: AI-Digital-Health-Twin
RAG-based legal AI assistant - fully deployed on AWS
Semantic retrieval pipeline over legal documents using AWS Lambda, DynamoDB, S3, and Amplify. Includes structured LLM outputs, caching layers, and fallback handling for edge cases. Built for users who can't access or afford legal counsel.
Stack: AWS Lambda · DynamoDB · S3 · Amplify · Claude API · RAG
Repo: Know-Your-Rights-AI
B2B lead qualification agent - LLM scoring + Notion CRM integration
AI agent that evaluates LinkedIn leads for ICP fit using contextual inference, signal extraction, and a weighted scoring system. Automates lead classification and personalized outreach generation for RevOps teams at SaaS companies.
Stack: OpenAI API · Prompt Engineering · Notion CRM · AI Automation
Repo: ICP-Signal-Scorer
Multi-model ML on combined NASA Kepler + TESS datasets
Trained and benchmarked Random Forest, XGBoost, and LightGBM on combined NASA Kepler + TESS transit photometry data for multi-class exoplanet signal classification. Includes feature importance analysis for interpretability.
Stack: Python · Scikit-learn · XGBoost · LightGBM · Jupyter
Repo: exodetect-project
Graph-based ML on the TrackML high-energy physics dataset
Reconstructed charged particle trajectories from detector hit data using the TrackML benchmark dataset - a problem from CERN-adjacent research. Demonstrates ML applied to scientific computing at a level rare for undergraduates.
Repo: Particle-Track-Reconstruction
📄 Harnessing Machine Learning in Fraud Detection: Techniques, Challenges, and Opportunities
Co-Author & Presenter · ICICC 2025
Topics: ensemble methods, anomaly detection, class imbalance, real-world deployment challenges
If you're building something at the intersection of AI and real-world access - Let's discuss more over coffee.
