GenAI Engineer obsessed with the gap between AI that demos and AI that ships. My sweet spot: multi-agent systems, RAG pipelines, and LLM workflows that handle real production load.
I started in Mechanical Engineering and pivoted hard into AI — context switching is kind of my thing. For 3+ years I've been turning AI research into shipped products on Google Cloud: code review agents, engineering intelligence systems, recommendation engines, and cost-anomaly detectors. If it involves LLMs, knowledge graphs, or agents doing real work under real load, I want to build it.
Production AI, not demos.
Building AI-powered developer tooling: a code review agent (Gemini + Neo4j knowledge graphs) deployed on Cloud Run, a multi-agent "Ask AI" assistant built with Google ADK and 90+ analytics tools via MCP, and LLM pipelines that improved review efficiency by ~40% across 500+ daily PRs — plus automated Sprint Summaries and DORA metrics.
Built a multi-tenant AI chatbot platform with RAG, LangChain, FAISS, and Google LLMs on Vertex AI; developed To&From, an LLM-powered gifting assistant; and shipped BillPulse, a GCP cost dashboard with anomaly detection and idle-resource cleanup.
Built a real-time speech emotion detection solution using NLP and OpenAI Whisper to enhance customer service interactions.
Developed GymX, a gym management app integrating ML algorithms and REST APIs for health tracking, member analytics, and operational efficiency.
Where the context switching began.
Each of these runs (or ran) in production with real users and real load.
An AI that reviews PRs using Gemini + a Neo4j knowledge graph of the entire codebase. It knows which files are risky, who owns what, and ranks suggestions by actual impact. Not just "looks good to me."
A multi-agent system where engineering leads ask plain-English questions and get answers backed by real sprint data, CI/CD metrics, and team analytics. 90+ live data tools, all orchestrated via MCP.
An LLM-powered assistant that matches gift attributes, occasion, and relationship context from user queries — custom prompts for precise attribute extraction, wired into product-matching APIs.
A GCP billing dashboard that detects cost anomalies before your cloud bill becomes a horror story. Idle-resource analysis and spend optimization included.
A multi-tenant RAG platform where companies train custom chatbots on their own PDFs, docs, web links, and sitemaps — FAISS vector search + Google LLMs on Vertex AI.
Classic Snake with a live top-3 leaderboard, speed-ups, and pause/resume — built with vanilla HTML, CSS, and JavaScript. Because engineers need fun side quests too.
Play now →The stack I reach for when it's time to ship.
Open to collaborations, interesting problems, and good conversations about production AI.
iampratik8@gmail.com