Ghaziabad, India — Independent AI Developer

I build AI that reads what bureaucracy writes.

Document intelligence, RAG systems, and multi-agent workflows — shipped into production for state governments and B2B teams, not prototyped for a demo day.

FILE REF: STONE QUARRY API
17+Document types parsed
95%+Extraction accuracy
3d → 30mReview time, per case

Case files

A working sample of systems built and deployed — for state governments, regulatory review, and B2B operators in the US and India.

FILE NO. 01 / GOVT Production
State Government — Citizen Services

Panchayati Raj Assistant

Multilingual RAG chatbot answering rural citizen queries on panchayat schemes, deployed in Odisha, West Bengal, and Uttar Pradesh with language-specific variants.

FastAPIAzure OpenAIChromaDBLangChainWhatsApp API
Hybrid MMR + similarity retrieval, Text-to-SQL over scheme data, live streaming answers over WhatsApp — no app install required.
FILE NO. 02 / REGULATORY Production
Environmental Clearance Review

Quarry Clearance Intelligence API

Parallel extraction and cross-validation across sand mine and environmental clearance proposal documents, with auto-generated deliberation minutes.

Azure Document IntelligenceAzure OpenAIFastAPIBlob Storage
17+ document types, 95%+ accuracy — review time cut from 3 days to 30 minutes.
FILE NO. 03 / GOVT Production
State Government — Scanned Records

Gujarat OCR Records Chatbot

RAG chatbot answering citizen queries directly from scanned Gujarati government PDFs, where standard OCR pipelines fail.

Surya OCRGoogle TranslateRAG
In-language answers from documents that were previously unsearchable.
FILE NO. 04 / GOVT Production
State Government — Registration Workflow

Palna Crèche Registration Assistant

Multi-agent system guiding applicants through crèche registration — document requirements, eligibility checks, and step sequencing.

LangGraphAzure OpenAI GPT-4oChromaDB
Orchestrated agent handoffs replace a manual, form-heavy intake process.
FILE NO. 05 / B2B — US Deployed
Freight & Logistics

Carrier Verification & Risk Scoring

Real-time carrier authority and safety verification for freight brokers, pulling live data from the FMCSA SAFER system with automated risk scoring.

FastAPIFMCSA SAFER APIRisk scoring engine
Replaces manual SAFER lookups brokers run before every load tender.
FILE NO. 06 / B2B — US Deployed
Home Services & Trades

AI Front Desk

Always-on AI receptionist for HVAC, plumbing, electrical, pest control, and garage door businesses — answers calls, books jobs, and qualifies leads.

Config-driven agent platformVoice + chat
Built to recover revenue lost to missed calls outside business hours.
Also shipped
Document fraud detection system Ration card RAG chatbot DataOps Text-to-SQL copilot Legal document intelligence pipeline B2B lead scoring & scraping tools

What I build

Three tracks, one underlying skill: turning messy, real-world documents and workflows into systems that run without a human re-reading everything by hand.

01

Document Intelligence

Extraction pipelines that turn scanned PDFs, handwritten forms, and inconsistent government or business paperwork into structured, verified data — at the accuracy a human reviewer would trust.

02

RAG & Conversational AI

Assistants that answer from your documents, in your users' language, on the channel they already use — web, WhatsApp, or voice — grounded so they don't guess.

03

Agentic Workflow Automation

Multi-agent systems and AI front-desk agents that carry out a real business process end to end — intake, verification, routing, follow-up — not just a chatbot on top of a website.

"Every project on this page is running somewhere right now — for a state government office, a freight broker, or a trades business — not sitting in a GitHub repo."
Based in — Ghaziabad, Uttar Pradesh, India
Background — IIT Roorkee
Focus — Document intelligence, RAG systems, LLM agents
Stack — FastAPI · Azure OpenAI · LangChain / LangGraph · ChromaDB
Clients — State government bodies (India) · B2B operators (US)
Also — Corporate trainer, several delivered cohorts

Start a case file

Send over what you're dealing with — the documents, the workflow, the volume — and I'll tell you plainly whether AI actually solves it.