I build production AI systems — from real-time voice AI agents to full-stack RAG apps. Final-year B.Tech at NIT Raipur, shipping real solutions with Python, FastAPI, LangChain & LangGraph.
from langgraph.graph import StateGraph
from fastapi import WebSocket
# 🎙️ Real-Time Voice AI Agent
agent = StateGraph(AgentState)
agent.add_node("classify", classify_intent)
agent.add_node("execute", run_tools)
agent.add_node("respond", generate_answer)
# 9 tools: RAG, booking, CRM, search...
tools = [search_knowledge, book_appointment,
lookup_customer, web_search, ...]
# Voice pipeline: Listen → Think → Speak
audio = groq_whisper(stream) # ~500ms
result = agent.invoke(audio)
edge_tts.speak(result) # ✓ Real-timeI'm a final-year B.Tech student at NIT Raipur (Class of 2026) who pivoted from Biotechnology into AI engineering — not just studying it, but building and shipping real systems.
My focus is on the Forward Deployed Engineering stack: taking complex AI capabilities (RAG, agentic workflows, LLM orchestration) and turning them into production-ready solutions that solve actual problems for real users.
At IIT BHU, I built data pipelines that achieved a 96.95% efficiency improvement. At NIT Raipur, I led a 300+ member team through high-stakes execution. I bring the same intensity to every line of code I write.
B.Tech · CPI 7.71/10
Data Analytics · Score 446
Data Pipelines · Python
Biotechnology · AIR 490
Open-source, production-grade voice AI agent with real-time WebSocket audio streaming. LangGraph-based stateful agent with 9 tools (RAG, booking CRUD, CRM, web search), Groq Whisper STT (~500ms), Gemini LLM, Edge-TTS, semantic caching, persistent memory, and Hinglish support. Real-time latency dashboard included.
Full-stack application that processes any YouTube video — downloads audio via yt-dlp, transcribes using Sarvam AI speech-to-text (30-sec chunks), and answers user questions with grounded LLM responses via Gemini. Built with FastAPI backend and FAISS vector indexing.
Single-file benchmark measuring retrieval speed vs. accuracy vs. silent-failure in RAG pipelines for voice AI agents. 42.3ms avg cold retrieval, 100% cache hit rate, 100% out-of-domain fallback across all configurations.
End-to-end Retrieval-Augmented Generation system tested across 50+ pages of real PDF documents. Chunks raw text, generates embeddings via Google Gemini, stores in ChromaDB, and retrieves relevant context to ground LLM responses with documented design decisions.
Object-oriented Python CLI application with JSON persistence, full CRUD operations, and explicit error handling for missing/corrupted data files. Degrades gracefully rather than crashing.
Normalized 3-table relational schema with foreign key constraints and a SQL window-function ranking algorithm to score listings by rent-to-distance ratio. Clean database architecture.
Relational database design + analytical SQL queries to rank agent performance, segment claim trends, and identify geographic risk patterns.
Analysis of personal Spotify streaming history using Python and Pandas.
Analysis of air quality data for Raipur, Chhattisgarh using Python.
I'm actively looking for internship opportunities in AI/ML, Forward Deployed Engineering, and Python development. If you think I'd be a good fit, let's talk.