Fullstack GenAI Engineer
Send a job offer directly to this candidate
I build production AI systems — the kind real users depend on daily, not demos.
For the past three years at First American I've shipped full-stack GenAI applications end to end: Python backends, RAG and agent layers, ETL pipelines over messy unstructured data, and the React interfaces teams actually work in. The results I'm proudest of are the boring operational ones — a document intelligence platform running 10,000+ documents a month at 92% accuracy for 1,000+ users, a LangGraph agent that took a two-hour manual workflow down to under four minutes, and an MCP server-based multimodal pipeline that cut a 30-minute expert review to five.
What I care about is the part most GenAI work skips: making it hold up. Labelled evaluation sets that decide whether a prompt or model change actually ships. Confidence gating so an uncertain answer goes to a human instead of quietly going out wrong. Retries, idempotency and rate-limit backoff so a long-running agent survives a bad afternoon of third-party APIs.
My stack: Python (FastAPI, Django), React, LangChain, LangGraph, MCP, RAG with FAISS/ChromaDB, PostgreSQL, MongoDB, Redis, Celery, Azure and Docker. I'm equally happy in a retrieval pipeline and a frontend component, which is usually what platform work needs.
Won my company-wide Prompt Engineering Hackathon against dedicated AI teams, and was named Q3 Top Performer and Star of the Quarter.
Open to GenAI / backend / full-stack engineering roles in Bengaluru or remote. Happy to talk shop about agents, evals, or what breaks in production.
Software Development Engineer — Full-Stack & GenAI
First American (India) Pvt. Ltd. · Full-time
Aug 2023 – Present · Bengaluru, Karnataka, India · On-site
Building production GenAI applications end to end — backend, data layer, LLM/agent layer and frontend — for 1,000+ users across 20+ automated workflows.
Led HiveSight, a production RAG platform (FAISS + embeddings, re-ranking, grounded generation) processing 10,000+ documents/month at 92% extraction accuracy for 1,000+ users
Built an autonomous multi-step LangGraph agent that reads inbound email and attachments, extracts seven critical fields and submits to a downstream API — a 2-hour manual workflow now completing in under 4 minutes (97% reduction)
Shipped an MCP server-based agentic pipeline using multimodal GPT-5 that reasons over scanned documents and diagram images through structured tool calls — expert review time down 85%+
Designed ETL and ingestion pipelines over unstructured documents: OCR and layout extraction, normalization, validation and persistence across SQL, NoSQL and vector stores
Engineered scalable Python services (Django, FastAPI) with distributed task execution via Celery/Celery Beat — retries, idempotency and rate-limit backoff for long-running LLM jobs
Built and maintained labelled evaluation sets per document type, using run-over-run accuracy to drive retrieval, chunking, prompt and model decisions
Worked directly with stakeholders from problem statement to production on 2–4 week cycles, then trained internal users and iterated on adoption
Awards: Company-wide Prompt Engineering Hackathon winner · Q3 Top Performer · Star of the Quarter
Bapuji Institute of Engineering and Technology (BIET), Davangere
Bachelor of Engineering (BE), Information Science and Engineering
Aug 2019 – May 2023 · Grade: 8.5 / 10
Optional line: Coursework in data structures and algorithms, DBMS, operating systems, computer networks and software engineering.
