Headline

AI Engineer | Production RAG & Multi-Agent Systems | Built tools used by 700+ users, IEEE & Springer author

About

I build AI systems that have to survive contact with real users, not just impress a demo audience. Right now that means production RAG pipelines, multi-agent workflows on LangGraph, and LLM apps that go from a rough idea to something a few hundred people rely on daily. The HR assistant I built on Microsoft Teams is a good example: it now handles 3,000+ monthly interactions for 400+ employees at Conneqtion Group and saves the HR team 45 to 60 hours a month. I got the response time from 7 seconds down to 4 and cut errors by 70% through prompt tuning, retrieval tuning, and a lot of time staring at Grafana dashboards. The problems I actually enjoy are the unglamorous ones. Getting retrieval to return the right document, not just the closest one. Teaching agents when to stop instead of confidently hallucinating an answer. Building evaluation harnesses that catch mistakes before a user does, not after. I've done this across a multimodal search system (78 to 82% top-1 accuracy across 500+ documents and 1,500+ images), a multi-agent pipeline that reads through 48,000+ records to speed up lead qualification, and a governed SQL agent that lets non-technical people query purchase orders and invoices without writing a line of SQL. Outside of client work, I fine-tune open models with LoRA and PEFT, and I've shipped a few of my own products, including an AI second brain called Dissect and a study-abroad assistant that's generated about 100 leads. Two of my papers on automated assessment are published in IEEE and Springer, which I'm still a little proud of. I'm looking for my next AI engineering role, ideally somewhere building agentic or RAG-based products in fintech or conversational commerce, where the stakes of getting it right are higher than a chatbot demo. If any of this sounds like a problem you're working on, message me here or email munnasuprathik79@gmail.com. I read everything.

Location

Bangalore Urban

AI stack

claude code

hermes agent

hugging face

kilocode

Browser Use

Humanloop

Work experience

AI Engineer at Conneqtion Group, Aug 2025 - Present · 1 yr 2 mos, Bengaluru, • Developed and maintained a production HR assistant on Microsoft Teams, enhancing efficiency for 400+ employees. • Achieved a 45–60 hour reduction in manual HR work per month through automation with Python, LangGraph, and LangChain. • Improved retrieval quality for a multimodal RAG search system, achieving 78–82% top-1 accuracy across extensive document and image datasets.

Intern at Conneqtion Group, Feb 2025 - Aug 2025 · 7 mos, Bengaluru, Karnataka, India

Python Developer at Freelance, Jan 2022 - Jan 2024 · 2 yrs 1 mo, Bengaluru

Education

Bachelor of Engineering, Computer Engineering, New Horizon College of Engineering

Public links

github: https://github.com/munnasuprathik

munnasuprathik.com: https://munnasuprathik.com/