role: Final-Year B.Tech IT Student
university: Parul Institute of Engineering & Technology, Parul University
location: India
focus: GenAI/LLM Engineering, Full-Stack MERN, Classical ML/CV
stack: LangChain, LangGraph, RAG, Multi-Agent Orchestration
vLLM, LiteLLM, FAISS, ChromaDB, React, Node, MongoDB
status: Open to internships & entry-level rolesI design LLM systems that behave like software, not demos — explicit multi-agent state machines, RAG pipelines, and retrieval infrastructure — paired with full-stack MERN builds shipping real-time features and clean architecture.
|
|
GenAI / LLM Engineering
Full-Stack Development
Core Languages, ML & CV
Tooling & Platforms
Proficiency Snapshot
LangGraph / Multi-Agent
RAG / Vector Retrieval
React / Node / MongoDB
Python / TensorFlow / OpenCV
Four agents — Planner, Researcher, Executor, Critic — run as an explicit LangGraph state machine over documents and live tools, rather than a loose prompt-chaining pipeline. Built for traceable execution and graceful fallback instead of silent hallucination.
- Explicit state-machine graph for inspectable, predictable execution
- Grounded document retrieval feeding the Researcher agent
- LiteLLM-backed model routing for provider fallback resilience
- PostgreSQL persistence via Alembic-managed migrations
Real-time speech recognition wired to LLM integration for hands-free system automation, reminders, and live info lookup. React/Vite frontend backed by a Node/Express/Socket.io server for live event streaming.
- Continuous speech-to-intent pipeline with low-latency response streaming
- Deployed with a split strategy — frontend on Netlify, backend on Railway
Ingests industrial PDFs, P&IDs, scanned forms, and logs into a unified, natural-language-queryable knowledge graph — turning messy, heterogeneous documentation into something an LLM can reason over.
A complete MERN hostel management system with role-based dashboards for Admin, Warden, and Student, plus Socket.io live updates across the app.
- Six modules: Room Allocation, Fee Management, Complaints, Notices, Visitor Logs, User Management
- JWT authentication with role-based access control
Classifies MRI scans as tumor-positive or normal via a CNN pipeline with curated train/validation splits.
A wide range of MERN apps demonstrating REST API design and modern JavaScript across many use cases.
Netflix Clone (Frontend) — React · Vite · Tailwind · TMDB API
Full frontend scaffold in place; building out MovieCard, ContentRow carousel, and TitleModal components, plus wiring in real Firebase Auth.
Conductor Hardening — LangGraph · LiteLLM · PostgreSQL
Taking Conductor from "working" to "production-styled" — end-to-end verification of the four-agent flow, live LiteLLM fallback wiring, and Alembic migrations against a live PostgreSQL instance.
The stats cards above render automatically via github-readme-stats — no setup needed beyond a public profile.
- Shipped 6+ portfolio-ready repos, spanning GenAI orchestration, RAG, MERN, and classical CV
- Conductor — went from prototype to an explicit 4-agent LangGraph state machine
- Nexus AI Assistant — real-time voice pipeline deployed across two hosts (Netlify + Railway)
- KnowledgeBrainAPI — unified ingestion for scanned and digital industrial documents
- 12+ MERN builds in the general portfolio, on top of two flagship deep-dives
- Currently pushing Conductor and the Netflix clone toward production polish
| Area | What I'm sharpening |
|---|---|
| Multi-Agent Systems | Deeper LangGraph orchestration patterns beyond simple chains |
| System Design | Fundamentals for designing agent state machines that scale |
| RAG Infrastructure | Retrieval quality, chunking strategy, hybrid search over FAISS/ChromaDB |
| Full-Stack Delivery | Shipping polished, real-time MERN features end-to-end |