Munim.ai

WhatsApp-first GST compliance co-pilot for Indian MSMEs

Traders forward invoices via WhatsApp. Munim extracts, validates, fraud-checks, and reconciles them automatically. CAs get a clean action-driven dashboard instead of a pile of paper.

---

Core USPs

1. WhatsApp-First — Zero App Download

• Traders interact entirely via WhatsApp in their language (Hindi, English, Marathi, Gujarati)
• Conversational onboarding in under 2 minutes
• Each trader gets a dedicated Munim email address for vendors who prefer email over WhatsApp

2. Multimodal AI Extraction (Gemini 2.5 Flash)

• Handles crumpled thermal receipts, handwritten bills, scanned PDFs, blurry photos
• Outputs structured JSON: supplier name, GSTIN, invoice number, date, line items, HSN codes, tax breakdown
• Low-confidence extractions are flagged for human review

3. Deterministic ITC Rules Engine — No LLM

• Pure rule-based GST Act §16 + §17(5) implementation
• Classifies every invoice into: CONFIRMED / FIXABLE_BLOCKED / AT_RISK / INELIGIBLE / FRAUD_FLAGGED
• 12+ blocked categories covered by HSN prefix and keyword matching (motor vehicles, accommodation, outdoor catering, health clubs, personal consumption, real estate, etc.)
• Zero hallucination risk — fully auditable logic

4. 6-Signal Fraud Scoring Engine (0–100)

Signal	What It Detects
GSTIN Age	New GSTIN (<180 days) issuing high-value invoices
Benford's Law	Unnatural leading-digit distribution (chi-squared test)
Sequential Invoice Numbers	Consecutive serials from same supplier — classic fake invoice pattern
Business Type Mismatch	GSTIN registration category contradicts invoice line items
Geographic Mismatch	Supplier state ≠ buyer state without IGST
Velocity Anomaly	Invoice amount >5× the supplier's historical average

Score ≥ 70 → FRAUD_FLAGGED. Score 40–69 → soft flag for CA review.

5. GSTR-2B Three-Pass Fuzzy Reconciliation

• Pass 1 — Exact: GSTIN + invoice number + date all match
• Pass 2 — Fuzzy: Levenshtein distance on invoice number (INV-001 vs INV001), ±2% amount tolerance, ±15-day date window
• Pass 3 — Amount + Date: Fallback when invoice number is ambiguous
• Unmatched invoices surface instantly in Action Queue with vendor contact options

6. Prioritized Action Queue

• All issues across all clients ranked: fraud flags → ITC-at-risk → fixable blocks
• Each item shows the exact reason, affected tax amount, and recommended fix
• One-click WhatsApp or email vendor warning sent directly from the dashboard

7. Supplier Health Monitoring

• Tracks each vendor's GSTR-1 filing consistency across months
• Flags chronically non-compliant suppliers before they become a problem

8. Email Invoice Ingestion (Cloudmailin)

• Traders share their dedicated Munim email with vendors
• Vendor emails PDFs → Cloudmailin webhook → same Gemini pipeline → auto-added to records

9. Auto-Generated Compliance Reports

• One-click PDF per trader per period
• Covers ITC summary, blocked amounts, at-risk credits, reconciliation status, supplier health

10. GST Portal Simulation

• Interactive IMS (Invoice Management System) + GSTR-3B auto-draft
• Populated from live backend data — context-aware per selected trader
• Mirrors the real GST portal UI for demo and training purposes

11. Real-Time Compliance Timeline

• Visual deadline calendar: GSTR-1, GSTR-2B upload, GSTR-3B filing dates
• Proactive WhatsApp reminders before each deadline

12. Multi-Tenant CA Dashboard

• One CA manages unlimited traders from a single login
• Instant client switching, fully isolated per trader data
• Built as a PWA — works on mobile without installation

---

Tech Stack

Layer	Technology
Backend	FastAPI, LangGraph, Python 3.12, Uvicorn
Frontend	Next.js 14 (App Router, Turbopack), Tailwind CSS
Database	Supabase (PostgreSQL + Row Level Security)
AI / LLM	Google Gemini 2.5 Flash (Vision + Text)
Messaging	Meta WhatsApp Cloud API
Email Ingestion	Cloudmailin
Cache / Sessions	Redis (Upstash)
GSTIN Validation	deepvue.tech API
Fuzzy Matching	python-Levenshtein
PDF Generation	WeasyPrint
Deployment	Railway (backend) + Vercel (frontend)

---

Architecture

Trader (WhatsApp / Email)
         │
         ▼
Meta Cloud API / Cloudmailin Webhook
         │
         ▼
FastAPI Backend (Railway)
         │
    LangGraph Pipeline
    ├── 1. Gemini Vision OCR → InvoiceJSON
    ├── 2. GSTIN Validator (deepvue.tech)
    ├── 3. HSN Validator (pgvector + Supabase)
    ├── 4. ITC Rules Engine   ← no LLM
    ├── 5. Fraud Scorer       ← no LLM
    └── 6. GSTR-2B Reconciler ← no LLM
         │
         ▼
Supabase PostgreSQL
    ├── CA Dashboard (Next.js / Vercel)
    │     ├── Action Queue
    │     ├── Supplier Health
    │     ├── ITC Timeline Chart
    │     ├── Reports Panel
    │     └── GST Simulation
    └── Redis (session state / conversation context)

---

Project Structure

munim-ai/
├── backend/
│   ├── app/
│   │   ├── api/
│   │   │   ├── webhook.py          # WhatsApp bot + onboarding
│   │   │   ├── dashboard.py        # CA dashboard endpoints
│   │   │   ├── gstr2b.py           # GSTR-2B upload + reconciliation
│   │   │   ├── reports.py          # PDF generation
│   │   │   ├── communications.py   # Vendor WhatsApp/email warnings
│   │   │   ├── email_webhook.py    # Cloudmailin ingestion
│   │   │   └── auth.py             # JWT auth
│   │   ├── domain/
│   │   │   ├── itc_engine.py       # GST §16/§17(5) rules
│   │   │   ├── fraud.py            # 6-signal fraud scorer
│   │   │   ├── reconciler.py       # 3-pass GSTR-2B reconciler
│   │   │   ├── hsn.py              # HSN validator
│   │   │   └── supplier_monitor.py # Supplier health scoring
│   │   ├── models/                 # Pydantic data models
│   │   └── services/               # Supabase, WhatsApp, Gemini clients
│   ├── schema.sql
│   └── requirements.txt
├── frontend/
│   ├── src/app/
│   │   ├── dashboard/              # CA main dashboard
│   │   ├── trader/                 # Trader PWA
│   │   └── components/             # Shared UI components
│   └── public/demo/                # GST simulation (standalone HTML/JS)
└── demo/                           # Symlinked for direct serving

---

API Reference

Endpoint	Method	Description
/api/v1/webhook	GET/POST	WhatsApp webhook (verify + message handling)
/api/v1/webhook/upload-invoice	POST	Direct upload from Trader PWA
/api/v1/email-webhook	POST	Cloudmailin inbound email → pipeline
/api/v1/dashboard/summary/{trader_id}	GET	ITC summary card data
/api/v1/dashboard/actions/{trader_id}	GET	Prioritized action queue
/api/v1/dashboard/actions/{id}/resolve	PATCH	Mark action resolved
/api/v1/dashboard/suppliers/{trader_id}	GET	Supplier health list
/api/v1/dashboard/invoices/{trader_id}	GET	Invoice records + filters
/api/v1/dashboard/itc-timeline/{trader_id}	GET	6-month ITC chart data
/api/v1/dashboard/reports/generate/{trader_id}	POST	Generate PDF report
/api/v1/gstr2b/upload-file/{trader_id}	POST	Upload GSTR-2B JSON
/api/v1/gstr2b/reconcile/{trader_id}	POST	Trigger reconciliation run
/api/v1/gstr2b/records/{trader_id}	GET	Fetch GSTR-2B records

---

Getting Started

Prerequisites

• Python 3.12+, Node.js 18+, Supabase project, Gemini API key, Redis, Meta WhatsApp credentials

Backend

cd backend
cp .env.example .env   # Fill: GEMINI_API_KEY, SUPABASE_URL, SUPABASE_SERVICE_KEY,
                        #       REDIS_URL, META_ACCESS_TOKEN, META_PHONE_NUMBER_ID,
                        #       CLOUDMAILIN_SECRET, DEEPVUE_API_KEY
pip install -r requirements.txt
uvicorn app.main:app --reload
# OR: docker-compose up -d

Frontend

cd frontend
cp .env.local.example .env.local   # Set NEXT_PUBLIC_API_URL=http://localhost:8000
npm install && npm run dev

• CA Dashboard → http://localhost:3000/dashboard
• Trader PWA → http://localhost:3000/trader
• GST Simulation → http://localhost:3000/demo

Database

Run backend/schema.sql in your Supabase SQL editor.

---

Production Deployment

Backend → Railway

1. New Project → Deploy from GitHub → Root Directory: backend/
2. Add Redis from Railway marketplace
3. Set all env vars in Railway → Variables
4. Auto-deploys on every push to main

Frontend → Vercel

1. New Project → Import repo → Root Directory: frontend/
2. Add env var: NEXT_PUBLIC_API_URL=https://your-railway-app.up.railway.app

WhatsApp Webhook

• URL: https://your-backend.up.railway.app/api/v1/webhook
• Verify Token: set in .env
• Subscribe to: messages

---

© 2026 Abhishek Saraf. All rights reserved. See LICENSE.