GenAI engineer. I build LLM systems that cite their sources
— RAG pipelines, agents, and the full stack around them.
Indore, India--:--:-- IST
Open to GenAI / full-stack roles
01 — About
I'm a curious and driven individual who thrives on solving meaningful
challenges through creativity, collaboration, and continuous learning.
Whether it's leading a team at a hackathon or exploring new technologies
on my own, I'm always looking for ways to make ideas come alive and
create real-world impact. I learn fastest by building, and I'd rather
ship something imperfect than theorise about it.
LLMs answer confidently even when they're wrong — which makes them hard to trust with private documents.
Approach
A full retrieval pipeline: documents are chunked and embedded into Supabase pgvector, candidates are re-ranked with Cohere to cut noise, and Gemini generates the answer — with inline citations pointing back to the exact source passage.
Outcome
A deployed, production-shaped system — FastAPI backend on Render, React frontend on Vercel. Ask a question, get an answer you can verify.
known limits — reranking is English-only for now; scanned PDFs still need an OCR pass.
Answers that trace back to the passage they came from.
An assistant that reads your inbox, your documents, and the web.
Problem
Email, documents and the open web each demand their own tab, their own search, their own context.
Approach
One FastAPI + LangChain service: Gmail over OAuth, where mail is summarized and replies are drafted but never sent without review; scraped pages chunked by size before embedding; FAISS vector search for RAG over everything it ingests — Gemini holds the conversation.
Outcome
A single assistant with memory across all three surfaces, controlled in natural language.
known limits — local-first by design; Gmail access requires per-user OAuth consent.
Mail, documents and the web, converging on one context.
Game controls assume a keyboard. Cameras see more than keys ever will.
Approach
MediaPipe tracks 21 hand landmarks per frame; a KNN model trained on gestures you record yourself maps them to keyboard input through a live transparent overlay — OpenCV in, simulated keys out.
Outcome
Camera-in, controls-out: hands-free Hill Climb Racing at real-time frame rates.
known limits — needs per-user gesture calibration; two-gesture vocabulary so far.
Intelligent Automation Specialist · EduSkills Foundation
First taste of automating the boring parts with ML.
Madhya Pradesh
2022 — 2026
B.Tech CSE (AI & ML) · ITM University
Google GenAI Academy alongside — five completed skill badges: Vertex AI prompt design, Gemini & Imagen applications, GenAI apps with Streamlit, multimodal RAG, generative AI with LLMs.
