AI Engineer: LLM Multi-Agent & RAG

Abhishek Kumar

AI Engineer LLM & Multi-Agent Systems RAG

+91-886******* *****************@*****.*** LinkedIn GitHub Professional Summary

AI Engineer specializing in production-grade Multi-Agent Systems and RAG pipelines, with a track record of cutting LLM token consumption by 20–30% and manual research time by 40% in live enterprise deployments. Proficient in LangGraph, Claude Agent SDK, LangChain, FastAPI, and vector databases (Qdrant), with hands-on experience building agentic workflows, hybrid memory architectures, and semantic search systems at scale. Focused on turning complex LLM capabilities into reliable, cost-efficient automation — from legal intelligence platforms to multimodal image retrieval systems.

Skills

AI & Agentic system LangGraph, LangChain, Claude Agent SDK, Agentic AI, Agent Orchestration, RAG, Qdrant, Vector Search, Semantic Search, MCP

LLM & Evaluation Claude 3.5 / 4.7, GPT-4o, Llama 3.2, Qwen, Gemma 3, Haiku 4.5, LLM Evaluation Backend & API FastAPI, Python, ASP.NET/C#, SQL

ML frameworks NumPy, Pandas, Matplotlib, Jupyter, OpenCV, OCR, Scikit-Learn Web Development: React.js, Next.js, Streamlit, Tailwind Other Tools: Docker, Redis, Git, GitHub, Tavily

Experience

Agentic AI Developer, KOMP Global Noida July 2026 – Present

• Built a multi-agent orchestration platform using LangGraph that plans and executes multi-step workflows - determining the right agents, tools, and execution order at runtime, with query understanding to turn user requests into actionable plans

• Integrated Model Context Protocol (MCP) to connect agents with external tools and enterprise applications, adding safeguards like input validation and prompt-injection checks to keep workflows reliable. Junior AI Engineer, S.S. Rana & Co. New Delhi March 2025 – July 2026

• Architected a domain-specific multi-agent system for IP law research using Claude Agent SDK, LangGraph, and Docker — automating legal query workflows and reducing attorney research time by 40%.

• Built a multi-agent email intelligence pipeline using Python, ASP.NET/C#, and Tavily to parse, classify, and structure 100+ client emails daily — replacing a fully manual process and saving 750+ hours annually.

• Architected an LLM-ready data ingestion pipeline using Python and BeautifulSoup to extract, clean, and embed website content into a structured vector knowledge base - improving RAG retrieval precision by 40%. Software Developer Intern, Origo Commodities India. Gurugram April 2024 – Feb 2025

• Developed a CNN-based handwriting recognition system using TensorFlow/Keras to digitize documents, achieving 90% accuracy - outperforming RNN (84%) and Logistic Regression (83%) baselines.

• Architected and deployed the real-time eAuction module for Origo eMandi using React.js and RESTful APIs, enabling live commodity bidding for farmers and traders across the platform. Projects

Multi-Agent Legal Research & Knowledge Intelligence Platform Claude Agent SDK · LangGraph · Qdrant · FastAPI · Redis · Docker

• Built a multi-agent legal research system using Claude Agent SDK, LangGraph, and FastAPI, integrating hybrid RAG, re-ranking, and session memory for accurate multi-turn queries over domain knowledge. Production-Ready Multimodal Semantic Image Retrieval System CLIP ViT-B/32 · DINO · Qdrant · FastAPI · Docker

• Built a multimodal image search API using CLIP ViT-B/32, DINO, Qdrant, and FastAPI over a 5k+ image corpus, enabling sub-second text-to-image and image-to-image retrieval through vector search. Education

Jagannath Community College, Delhi 2022 – 2024

MCA (Master of Computer Application)

Contact this candidate