Skip to content

Transaction AI - Intelligent Financial Categorization

Welcome to the Transaction AI documentation! This system provides enterprise-grade financial transaction categorization with 98.7% accuracy, complete privacy, and full transparency.

What is Transaction AI?

Transaction AI is an open-source, privacy-first system that automatically categorizes financial transactions using a novel ensemble approach combining:

  • MCC (Merchant Category Codes) - Industry-standard merchant classification
  • Rule-based matching - Pattern recognition and keyword matching
  • Machine Learning - SetFit embedding-based classification
  • LLM fallback - Llama 3.1 for ambiguous cases

Key Features

🔒 Privacy First

  • 100% on-premise - No external API calls
  • Zero data sharing - All processing happens locally
  • GDPR compliant - Full data sovereignty

🎯 High Accuracy

  • 98.7% overall accuracy across all categories
  • 85% auto-accept rate - Minimal manual review needed
  • Ensemble voting - Multiple methods for reliability

💡 Transparent & Explainable

  • Full decision transparency - See why each prediction was made
  • 5-level explainability - From simple to detailed explanations
  • Open-source code - Complete transparency (MIT License)

⚡ Production Ready

  • Docker deployment - Quick setup with docker-compose
  • RESTful API - Easy integration
  • Real-time processing - <100ms response time
  • Monitoring included - Prometheus & Grafana dashboards

Quick Start

# Clone the repository
git clone https://github.com/Rahul1269227/transaction-ai.git
cd transaction-ai

# Start the system
docker-compose up -d

# Test the API
curl -X POST http://localhost:8000/categorize \
  -H "Content-Type: application/json" \
  -d '{"text": "STARBUCKS COFFEE $4.50"}'

Use Cases

  • Personal Finance - Automatic expense tracking and budgeting
  • SMB Accounting - Automated bookkeeping for small businesses
  • Banking Apps - Transaction categorization for mobile banking
  • Tax Preparation - Categorize transactions for tax filing
  • Research - Financial behavior analysis and economic research

Why Choose Transaction AI?

Feature Transaction AI Commercial APIs
Cost Free (self-hosted) $24K-300K/year
Privacy 100% on-premise Data sent to cloud
Accuracy 98.7% 85-95%
Customization Full control Limited
Transparency Open-source Black box
Rate Limits Unlimited 100-1K req/min

Architecture Overview

┌──────────────────────────────────────────────────────────┐
│                  TRANSACTION INPUT                        │
│              "STARBUCKS COFFEE $4.50"                     │
└─────────────────┬────────────────────────────────────────┘
┌──────────────────────────────────────────────────────────┐
│               ENSEMBLE CATEGORIZER                        │
│  ┌───────────┬────────────┬──────────┬────────────┐     │
│  │   MCC     │   Rules    │    ML    │    LLM     │     │
│  │ Lookup    │  Matching  │ SetFit   │  Llama 3.1 │     │
│  │  95%      │    90%     │   88%    │   (skip)   │     │
│  └─────┬─────┴──────┬─────┴─────┬────┴──────┬─────┘     │
│        │            │           │           │           │
│        └────────────┴───────────┴───────────┘           │
│                      │                                   │
│              ENSEMBLE VOTING                             │
│          (Unanimous: 3/3 agree)                         │
└─────────────────┬────────────────────────────────────────┘
┌──────────────────────────────────────────────────────────┐
│                  FINAL PREDICTION                         │
│  Category: food_dining                                    │
│  Confidence: 95%                                          │
│  Method: ensemble_unanimous                               │
│  Requires Review: false                                   │
└──────────────────────────────────────────────────────────┘

Documentation Structure

This documentation is organized into three main sections:

📚 Core Documentation

Foundational information about the system architecture, data strategy, and responsible AI practices.

🚀 Innovation & Features

Details about novel technical approaches, explainability, continuous learning, and bias mitigation.

📊 Impact & Metrics

Business impact, scalability metrics, evaluation results, and broader societal impact.

Community & Support

License

Transaction AI is released under the MIT License, enabling anyone - from solo developers to Fortune 500 companies - to deploy enterprise-grade AI without vendor lock-in or prohibitive costs.

Next Steps