AI Graduate Student RAG & Agentic Systems
Developed Sourava, a privacy-focused iOS voice-memory assistant featuring on-device speech transcription, semantic search, vector retrieval, and RAG-based conversational AI for intelligent memory recall and productivity workflows.
Experienced in building retrieval-augmented generation (RAG) pipelines, autonomous agent workflows, and conversational AI systems using Python, OpenAI APIs, Firebase, and MongoDB. Skilled in designing scalable AI architectures involving vector embeddings, semantic search, NLP pipelines, geospatial retrieval, and cloud-based automation.
Built AI-powered platforms including ride-sharing systems, event networking assistants, recommendation engines, and edge AI applications deployed on Raspberry Pi devices. Passionate about developing production-ready, context-aware AI systems that combine intelligent automation with real-world usability.
iOS voice-memory app with on-device Whisper transcription, semantic indexing (Core ML + vector store), and RAG Q&A over personal recordings. Built with Swift/SwiftUI; features include scheduling tools, lecture-to-PDF Classroom mode, and speaker-aware transcripts. Privacy-oriented, local-first design with optional OpenAI integration.
Ibhood is a local community platform that helps neighbors connect for everyday needs — ridesharing, short-term stays, hyperlocal posts (HAP), and neighborhood offers — in one simple app. Launching in South Florida and available on the App Store, it is built as a community hub rather than a carrier, hotel, or employer: people arrange rides and stays directly with each other, backed by messaging, ratings, and location-aware discovery.
At the center of discovery is Global Ask, powered by an Ask RAG pipeline. Users ask plain-language questions like “Any rides near FAU tonight?” or “What’s happening in Fort Lauderdale?” The system detects intent, retrieves relevant community content — including vector/geo search over HAP posts and structured search across rides and stays — then returns a short, grounded answer with tappable source cards. That retrieval-augmented approach keeps answers tied to real listings and posts instead of free-floating AI guesses.
Together, HAP, Rideshare, Stay, Offers, messaging, and Ask RAG make Ibhood a practical neighborhood toolkit: browse on a map, post what you need, chat in-app, and search across the community in natural language when you just want a quick, trusted answer.
View on the App StoreBuilt a multi-agent, AI-driven ride-sharing system with dual Telegram bots enabling conversational ride booking through OpenAI-powered NLP. Engineered a retrieval-augmented geospatial matching pipeline using MongoDB 2dsphere and Firebase Cloud Tasks, integrating RAG-based context retrieval and autonomous agent coordination for real-time scheduling and dynamic ride optimization.
View Live Project
Developed an intelligent event networking bot powered by multi-agent collaboration and RAG pipelines to connect attendees at hackathons or conferences based on shared interests and extracted resume skills. Implemented seamless onboarding via event-specific QR codes and unique event IDs, enabling real-time participant matching, AI-driven team formation, and context-aware engagement throughout the event lifecycle.
Built a sandwich recommendation engine using voice input and daily retraining for improved accuracy.
Deployed a model on a Raspberry Pi to classify 200+ drone types from camera input using YOLO.
Created a disease prediction chatbot with 88% accuracy and published research on fine-tuned BERT models.
View IEEE PublicationMaster of Science, Artificial Intelligence
Open to collaborations, research, and AI engineering opportunities.
