AI engineer in Gurgaon, India. I build production agentic AI — RAG pipelines, multi-agent orchestration, MCP servers, LLM evals — and keep them running for enterprise clients (Mitsubishi Fuso / Daimler Truck, Rohto Pharmaceutical). M.Tech, IIT Kharagpur.
Currently: Consultant, AI Engineer @ Cubastion Consulting. Open to relocation (Tokyo / global).
- TIC — production RAG chatbot for 3,000+ engineers across 200+ Mitsubishi Fuso service centers in Japan. Owned end to end: ingestion over 15 years of troubleshooting cases → retrieval tuning → AKS deployment and monitoring. Re-architected serving and cut cloud cost 62% (¥3.2M → ¥1.2M/mo).
- MCP server, 25+ chat-native analytics tools — LLM clients answer cross-domain supply-chain questions with interactive panels (KPI cards, SVG genealogy graphs, CSV export), results identical to the BI portal.
- QMPQR — multi-agent warranty analytics over 2M+ row datasets; SQL-programming agents replaced a manual weekly quality report.
- SlideX — AI deck generation (FastAPI + React 19) with a visual-fit quality loop: every slide renders in headless Chromium, gets measured for overflow/clipping/overlap, and is repaired or rebuilt deterministically before it ships.
- BYOM coding agent, 20+ tools for PR generation, review, debugging, and test execution — used daily by internal teams.
| dbvault | Multi-database backup utility — Fernet encryption, S3/Azure upload, async execution, retry logic |
| localmind-broswer-rag | Fully client-side RAG in the browser — embedding, retrieval, and inference with no server round-trip |
| coma-context-framework | OS-inspired context/memory allocator for long-horizon LLM agents under token budgets |
| hybrid-music-recommender-agentic-ai | Agentic recommender with RL and self-healing |
| ai_agents | CLI coding agents |
| tiny-os-y · CHIP8-EMU · voice-mascot | A minimal OS, a CHIP-8 emulator, and an Arduino/ESP8266 voice assistant — for when I miss the metal |
- LocalMind: Browser-Native RAG System (2026) — 10.5281/zenodo.20578399
- COMA: Context Memory Allocator — A Memory Management Framework for LLM Agents (2026)
- AI Podcast Generation: Automated Multi-Speaker Audio Content Pipeline (2025)
Python SQL C++ JavaScript/TypeScript Swift Dart
LangGraph LangChain MCP CrewAI Azure OpenAI RAG LLM evals
FastAPI Flask Express WebSockets SSE
Azure AKS Kubernetes Docker GitHub Actions Nginx
MySQL PostgreSQL Redis MongoDB Neo4j
日本語勉強中 (JLPT N5) · English · हिन्दी · বাংলা