Building production-grade generative AI — retrieval-augmented generation, LLM orchestration, and agentic systems with LangChain, LangGraph & FastAPI. From embeddings to deployment.
I'm an AI/ML undergraduate at VVIT specializing in production-ready generative AI applications — from data ingestion and embeddings to full RAG pipelines and agentic orchestration.
My work spans LangChain and LangGraph for multi-step reasoning, vector databases like Qdrant, FAISS and Chroma for retrieval, and FastAPI backends that hold it all together — with a strong foundation in DSA, machine learning, and backend systems engineering.
The center is the engineer; every dendrite is a tool that carries a signal into production.
Designed and implemented efficient telecom traffic data storage using Berkeley DB, optimizing query performance for large-scale network topology and traffic matrices. Built structured retrieval mechanisms improving data access efficiency for enterprise-scale datasets.
Coordinated with speakers and managed on-ground logistics, ensuring smooth execution of the event and strong audience engagement, working within a cross-functional team.
Enterprise-grade RAG system built on LangGraph — agentic workflows with multi-step reasoning, conversation memory, and history-aware query planning. NeMo Guardrails block off-topic queries, prompt injection and jailbreak attempts. Portkey Gateway auto-fails over across Groq API keys for high availability. Retrieval runs on Qdrant Cloud with 3072-dim Gemini embeddings and FlashRank re-ranking.
A retrieval-augmented generation chatbot answering questions grounded in uploaded documents. Multi-document collections indexed with semantic chunking and embeddings for accurate, context-aware retrieval and reduced hallucinations.
A scalable full-stack messaging platform with real-time communication, disappearing messages that cut storage consumption by roughly 40%, and secure authentication.
Voices come from your browser/device — Chrome and Edge usually include the most natural-sounding ones. Pick whichever sounds best to you.