RAJKOT, INDIA/// BUILDING AGENTIC AI SYSTEMS

Mohit
Baraiya.

AI/ML engineer building agentic RAG systems — hybrid retrieval, graph reasoning, and LangGraph pipelines, shipped as real deployed products, not notebooks.

SCROLL
LANGGRAPHAGENTIC RAGGRAPH RAGPINECONEFASTAPISUPABASEHYBRID SEARCHGROQ · LLAMA 3TAVILYNETWORKX

Systems that retrieve, reason, and actually ship.

AI/ML Engineering student at Government Engineering College Rajkot, focused on production LLM systems — hybrid retrieval RAG, graph-based RAG, and agentic pipelines built with LangGraph.

I've shipped two full-stack AI applications end to end — backend to deployment — and I'm currently building an agentic RAG system combining vector, keyword, and graph retrieval behind a quantitative evaluation layer.

Before this: a Flutter Developer internship building real-time infrastructure with Firebase — where the full-stack instinct started.

Projects worth reading the source for.

01

GraphAgenticRAG

IN PROGRESS
Agentic RAG · Hybrid + Graph Retrieval

A production agentic RAG system combining hybrid search (Pinecone dense + BM25 sparse) with graph-based retrieval (NetworkX) and live web search (Tavily) for multi-source Q&A.

  • LangGraph agent state machine with a verifier / re-evaluation loop that checks retrieved context before response generation.
  • FastAPI backend with SSE streaming and Supabase Auth (JWT via JWKS, ES256) for secure real-time multi-user sessions.
  • Building a RAGAS evaluation layer to measure retrieval precision and recall.
PythonLangGraphPineconeBM25NetworkXTavilyFastAPI · SSE
02

Blueprint Agent

LIVE
LangGraph Agentic Pipeline

An 8-node LangGraph pipeline that turns a vague project idea into a downloadable PDF blueprint — market validation, system architecture, tech stack, and a phase-wise build plan.

  • Human-in-the-Loop via LangGraph's interrupt() — pauses mid-execution for clarification, resumes after the user answers.
  • Tavily web search inside a validation node checks competitors and market fit before architecture generation.
  • Skill-aware stack recommendations and a final PDF export with an embedded SVG architecture diagram.
PythonFastAPILangGraphGroq · LLaMA 3TavilyRender
03

SourceMind AI

LIVE
Hybrid RAG Chatbot

Hybrid RAG chatbot combining dense HuggingFace embeddings with BM25 sparse vectors on Pinecone, using parent-child semantic chunking (~390 indexed chunks) for PDF and YouTube transcript Q&A. Full session isolation with conversation memory in Supabase.

PythonLangChainPineconeHuggingFaceGroq · LLaMA 3Supabase

What I build with.

Generative AI

RAGHybrid RAGGraph RAGAgentic AILangGraphHITLTool CallingPineconeGroq (LLaMA 3)

ML & Data

EDARegressionDecision TreesRandom ForestK-MeansEnsemble Learning

Deep Learning

ANNCNNRNNLSTMGRUTransformersKerasTensorFlow

Languages

PythonSQL

Frameworks & DB

FastAPILangChainLangGraphSupabasePinecone

Tools

RenderGitHub PagesGitVS CodeJupyter

Experience.

Jan 2025 — Jun 2025

Flutter Developer Intern

Fuerte Developers · Rajkot, India
  • Built Aapka Vyapar, a cross-platform business management app for small shopkeepers covering sales, stock, purchases, and customer records.
  • Integrated Firebase authentication, Firestore real-time database, and push notifications; handled API integration and state management.

Education.

2025 — Present

B.E. in AI and Data Science

Government Engineering College Rajkot
Rajkot, India · Currently in 5th Semester
Completed 2025

Diploma in Computer Science & Engineering

R.K. University School of Diploma Studies
Rajkot, India · SGPA: 8.90

Let's build something.

mohitbaraiya761@gmail.com