Resume

• Designed and built an end-to-end RAG pipeline: document ingestion, chunking, embedding generation, and vector retrieval.
• Built scalable REST APIs with FastAPI for RAG inference — query processing, retrieval, and LLM orchestration.
• Developed a React frontend for real-time querying and response display.

• Proposed a novel feature using the modulation spectrogram at vowel onset points for fake speech detection — outperforming existing baselines on standard datasets.
• Explored SFF and LFCC feature extraction with ML classifiers for spoofing detection.
• Built autoencoders to compress features from 128 → 16 dimensions, improving classification accuracy.

Deepfake Detection using Blockchain

Web3-enabled detection platform: CNN + Vision Transformer models with results verified on Ethereum and stored on IPFS.

Production RAG Pipeline (Dvara ML)

End-to-end document ingestion → embedding → retrieval → LLM, served via FastAPI with a React frontend, containerized with Docker.

nCAKES — P2P Video Streaming System

BitTorrent-inspired peer-to-peer video chunk streaming over TCP/IP with a custom JSON application-layer protocol.

Natural Language → MongoDB Query Chatbot

LangGraph agent that converts plain-English questions into executable MongoDB queries over MongoDB Atlas.

Sentiment Analysis on 1.6M Tweets

Classical NLP sentiment classifier trained on 1.6 million tweets using NLTK, gensim, and scikit-learn.

Banking Churn Prediction

Gradient-boosted XGBoost classifier predicting customer churn with 85% accuracy for proactive retention.