AI/ML Engineer — Production Agentic AI
I build LangGraph multi-agent systems, hybrid RAG pipelines, and LLM-powered enterprise applications that actually work in production — not just in demos.
I'm an AI/ML Engineer at Vishleshan AI Solutions with 2.5+ years of production experience. I specialize in building systems that go beyond prototypes — handling real traffic, real edge cases, and real business constraints.
Currently pursuing M.Tech in AI & Data Science from IIT Patna alongside full-time work. My focus areas are LangGraph-based agentic systems, hybrid RAG architectures, and multilingual NLP for Indian enterprise use cases.
I've diagnosed silent LangChain batching bugs, fixed BM25 indexing on structured payloads, and dropped embedding latency from 15 seconds to 0.1 seconds by switching to local inference — the kind of problems that only appear when you build for production.
LangGraph Supervisor Agents orchestrating specialised subagents with stateful multi-step reasoning.
Hybrid retrieval with RRF reranking, RAGAS evaluation, and sub-400ms latency at enterprise scale.
Hindi, Hinglish, and English pipelines for Indian enterprise use cases — search, translation, extraction.
Pursuing M.Tech in Artificial Intelligence & Data Science from IIT Patna (2026–2028).
Production systems with real users, real traffic, real constraints.
LangGraph Supervisor Agent orchestrating a RAG Agent and Web Search Agent for dynamic multi-step query resolution. Supports PDF, DOCX, PPT, CSV, Excel, and websites with zero context loss — Unstructured.io element chunking, pdfplumber table extraction, and S3 image retrieval for truly multimodal inference.
Hybrid RAG pipeline for a major vehicle manufacturer. Dense semantic + BM25 sparse retrieval with RRF reranking. Multilingual Hindi/Hinglish/English support with intent classification and multi-turn conversation via Redis session history.
JS embed SDK that translates any website in real-time using Sarvam AI. Single script tag injection — zero code changes on client side. Stale-while-revalidate Redis caching for near-zero perceived latency. Supports Hindi, Marathi, Tamil, and 10+ Indian languages.
Production NER pipeline for identification document extraction at JSW GBS. 92% F1 score across 150+ fields processing 5,000+ documents per day. Projected to eliminate 11 FTEs annually through automated document understanding.
Open to remote and hybrid AI Engineer roles. If you're building something ambitious with LangGraph, RAG, or agentic AI — let's talk.