Open to opportunities

Harshit Narula

@harshitnarula

Message

Full Stack AI Engineer building production LLM, RAG, and multimodal agent systems.

India

Message

What I'm looking for

I’m looking to build enterprise-grade LLM, RAG, and agentic systems that ship to production—leveraging scalable architectures and measurable performance gains, collaborating in agile teams, and owning end-to-end delivery from pipelines to APIs.

I’m a Full Stack AI Engineer with 1.5+ years of experience taking LLM applications from concept to production—especially retrieval-augmented generation (RAG) and multi-modal AI agent systems.

At HOI (GMR Group), I architected an enterprise multi-modal AI agent (HOIAI Agent) using LangGraph and LangChain, automating airport hospitality workflows and reducing manual processing by 70%.

I’ve also built production-grade RAG over 500,000+ documents (FAISS/ChromaDB + OpenAI/Anthropic embeddings), improved task accuracy by 35%, cut response latency by 60% with FastAPI + Pinecone + RabbitMQ, and optimized 200+ PostgreSQL queries to reduce execution time by 40%.

Experience

Work history, roles, and key accomplishments

Current

Full Stack AI Engineer

Current

HOI (Gmr Group)

Jan 2025 - Present (1 year 5 months)

Architected an enterprise multi-modal AI agent system using LangGraph/LangChain and RAG, reducing manual airport hospitality processing by 70%. Built a production RAG pipeline over 500,000+ documents with FAISS/ChromaDB and LLM fine-tuning (PEFT/LoRA), improving task accuracy by 35% and cutting response latency by 60%.

LangGraph LangChain Retrieval Augmented Generation (RAG)Faiss ChromaDB PEFT LoRA fastAPI RabbitMQ PostgreSQL AWS Lang Chain

Full Stack Engineer Intern

HOI (Gmr Group)

Aug 2024 - Jan 2025 (5 months)

Developed full-stack reservation, user management, and reporting features using React.js, NestJS, and PostgreSQL, serving thousands of daily active users. Automated SFTP-to-AWS ingestion pipelines and integrated RabbitMQ for asynchronous event processing, reducing pipeline delays by 60%.