I'm building machine learning and scientific-computing projects at IIT Kanpur, from a RAG-based PDF analyzer using Gemini, Cohere, FAISS, Streamlit, and Mermaid.js to automated visual indexing workflows.
For an IITK EEA project, I developed PyTorch Geometric models for Bitcoin fraud detection on the Elliptic dataset. My GraphSAGE ensemble achieved 0.7607 Macro-F1 and 0.0221 FPR while using 43% fewer features than an XGBoost baseline.
I'm also benchmarking Pan-Human Azimuth cell-type annotations across eight human tissue scRNA-seq datasets, achieving blood-tissue hF1 of 0.979. Alongside this work, I apply Bayesian MCMC methods to uncertainty analysis in soil transformation kinetics.
