🧠 MatchMind: Autonomous AI Finance Controller & Multi-Source Reconciliation Agent

Built for Razorpay Buildathon Hackathon
An autonomous financial reconciliation engine that ingests Bank Feeds, General Ledgers, and Invoices/Credit Notes, performs 3-way multi-tier reconciliation, resolves Indian payment gateway fees (Razorpay MDR & TDS), guards against duplicate collisions, and models 30/60/90-day forward cash runway.

---

📖 Comprehensive Documentation

For a deep dive into system architecture, multi-tier decisioning, Indian fintech logic, cash runway modeling, and full data flow, read PROJECT_DOCUMENTATION.md.

Live LINK -

https://ai-finance-controller-upla.onrender.com/

🏛️ System Architecture

How it works (5-second version): financial records flow in from the left → the Web Dashboard + REST API accept and scope them → the Reconciliation Orchestrator runs a 7-step workflow (ingest → normalize → compare → candidates → evaluate → exceptions → synthesize) → the Matching Engine tries deterministic tiers first (Exact → Batch → Fuzzy) and sends only ambiguous cases to Google Gemini → Policy & Validation gates every decision → results fan out to Reconciled Results, Exception Queue, and Reports & Audit, all retained in Session & History Storage.

• Deterministic first, AI last — AI never owns the pipeline; it only sees cases rules can't decide, bounded by confidence thresholds, approval rules and safety checks.
• Clear boundaries — users, data sources and Gemini sit outside the MatchMind application boundary; everything inside is the system we built and demo.

---

⚡ Core Superpowers

1. 🇮🇳 Indian Fintech & Razorpay Native:

   • Denominated in Indian Rupees (INR / ₹) with Lakhs/Crores formatting.
   • Razorpay Standard MDR Solver: Accurately matches gross invoices with net bank settlements after deducting 2.0% MDR + 18% GST (2.36% total fee).
   • TDS Compliance: Resolves Section 194-O (1%), Section 194C (2%), and Section 194J (10%) deductions.
   • Razorpay Payouts: Detects IMPS (₹5.90) and NEFT/RTGS (₹11.80) payout fees.

2. 🛡️ 5-Tier Decisioning Engine:

   • Pass 0: Phantom Net-Zero UPI & Auto-Reversal Filter (prevents fake matches on failed transactions).
   • Tier 1: Deterministic Exact & Precision Paisa Round-off Rule ($\le ₹0.99$ GST rounding).
   • Tier 1.5: Credit Notes & Debit Notes (aggregates sales returns into net bank payouts).
   • Tier 2: Multi-Signal Fuzzy Matcher with 30-day adaptive terms window.
   • Tier 3: Autonomous AI Reasoner (Gemini 2.5 Flash) with Indistinguishable Twin Collision Guard.

3. 📈 30/60/90-Day Forward Cash Runway Forecaster:

   • 4-Step Forward Liquidity Trajectory (Today, +30d, +60d, +90d).
   • Accounts Receivable (AR) vs Accounts Payable (AP) Aging Buckets (0-30d, 31-60d, 61-90d+).
   • Dynamic runway calculation and monthly net burn rate monitoring.

4. 💻 Full-Stack Interactive UI:

   • Built with React 19, Tailwind CSS v4, and Lucide Icons.
   • High-contrast financial dark mode design token system with semantic color mapping.
   • Real-time 3-way transaction grid, interactive runway trajectory chart, and exception triage queue.

---

💻 Getting Started (Local Development)

1. Install Dependencies

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

2. Launch Local Server

python server.py

Open http://127.0.0.1:8000 in your browser.

3. Run Benchmark CLI

python run_agent.py --seed 42 --llm auto

---

🚀 Cloud Run / Docker Deployment

The repo includes a production multi-stage Dockerfile (Vite build + FastAPI backend):

docker build -t matchmind .
docker run -p 8000:8000 -e GEMINI_API_KEY="your_api_key" matchmind