🛡️ ComplianceAI — AI-Powered Anti-Money Laundering Platform

Intelligent transaction monitoring using Machine Learning, RAG, and Gemini AI — with Human-in-the-Loop review.

---

📌 Problem Statement

Every year, over $2 trillion is laundered globally. Financial institutions spend $274 billion on compliance, yet 95% of money laundering goes undetected. Current compliance systems rely on static, rule-based thresholds (e.g., "flag transactions over $10,000") that criminals easily evade through structuring and smurfing. When alerts do trigger, human analysts spend 30–45 minutes per case manually reviewing transactions, cross-referencing policies, and writing reports — creating massive backlogs and fatigue.

---

💡 Our Innovation

ComplianceAI replaces the entire manual compliance pipeline with an end-to-end AI-powered system:

🌳 ML-Based Detection (Not Rules)

A Random Forest classifier trained on 15 engineered features — including transaction velocity, cross-currency flags, implied exchange rates, and temporal patterns. Achieves 99% accuracy on the IBM AML dataset, catching patterns that rule-based systems miss.

🧠 RAG-Grounded AI Explanations

Every flagged transaction gets a professional compliance explanation from Google Gemini 2.5 Flash, grounded in real AML policy documents via Retrieval-Augmented Generation (RAG) using ChromaDB. No hallucinations — every explanation is backed by actual regulatory policy.

👤 Human-in-the-Loop Review

AI recommends, humans decide. Compliance officers can Resolve or Escalate flagged violations with notes, creating a complete audit trail. The AI reduces analyst workload from 30 minutes to a quick review.

⚡ Distributed Processing

Celery task workers with MongoDB Atlas as the broker enable asynchronous batch processing. Each transaction is analyzed in ~3-5 seconds — ML prediction, policy retrieval, and LLM explanation — all in the background.

---

🏗️ Architecture

┌──────────────┐     REST API        ┌──────────────────┐
│  React + TW  │ ←─────────────────→ │  FastAPI Server  │
│   Frontend   │                     │  (Python)        │
└──────────────┘                     └────────┬─────────┘
                                              │
                              .delay()        │
                           (queue tasks)      ↓
                                     ┌──────────────────┐
                                     │  MongoDB Atlas   │
                                     │  (DB + Broker)   │
                                     └────────┬─────────┘
                                              │
                                    picks up  │  messages
                                              ↓
                                     ┌──────────────────┐
                                     │  Celery Worker   │
                                     │                  │
                                     │  1. ML Predict   │
                                     │  2. RAG Retrieve │
                                     │  3. Gemini API   │
                                     │  4. Save Results │
                                     └──────────────────┘

---

🧰 Tech Stack

Layer	Technology
Frontend	React, Tailwind CSS, Recharts, Lucide Icons
Backend	FastAPI (Python)
ML Model	Random Forest — scikit-learn (150 trees, 15 features)
LLM	Google Gemini 2.5 Flash
RAG	ChromaDB + Sentence-Transformers + Gemini Embeddings
Task Queue	Celery (solo pool)
Database	MongoDB Atlas

---

📊 ML Model Details

• Dataset: IBM AML Transaction Dataset (HI-Small_Patterns.txt)
• Algorithm: Random Forest Classifier (150 estimators, max depth 20)
• Accuracy: 99.07% | Precision: 0.99 / 0.98 | Recall: 0.99 / 0.98

Feature Engineering (15 features)

Feature	Type
Log_Amount_Paid / Log_Amount_Received	Log-transformed amounts
Implied_Exchange_Rate	Received / Paid ratio
Is_Cross_Currency	Boolean — different currencies
Same_Bank_Transfer	Boolean — same bank sender/receiver
Sender_Tx_Frequency / Receiver_Tx_Frequency	Account velocity
DayOfWeek, Is_Weekend, Time_of_Day	Temporal features
From_Bank, To_Bank, Payment_Currency, Receiving_Currency, Payment_Format	Label-encoded categoricals

---

📂 Project Structure

ComplianceAI/
├── backend/
│   ├── agents/            # AI agents (monitoring, detection, explanation, RAG, reporting)
│   ├── api/routes/        # FastAPI endpoints (violations, agents, predictions, RAG)
│   ├── config/            # Settings and environment config
│   ├── core/              # Celery app and startup logic
│   ├── datasets/          # Transaction data files
│   ├── ml/                # Random Forest model + predictor
│   ├── models/            # Pydantic schemas
│   ├── notebooks/         # Training notebooks
│   ├── rag/               # Vector store + embedding model
│   ├── scripts/           # Data loading utilities
│   ├── security/          # Encryption helpers
│   ├── database.py        # MongoDB connection layer
│   ├── tasks.py           # Celery task definitions
│   ├── main.py            # FastAPI application entry
│   └── requirements.txt   # Python dependencies
├── secure-ai-dashboard/
│   └── src/
│       ├── pages/         # React pages (Dashboard, Agents, Violations, etc.)
│       ├── components/    # Reusable UI components
│       └── services/      # API service layer
├── .gitignore
└── README.md

---

🚀 Steps to Run the Project

Prerequisites

• Python 3.10+
• Node.js 18+
• MongoDB Atlas account (or local MongoDB)
• Google Gemini API Key

1. Clone the Repository

git clone https://github.com/abhay-2108/ComplainceAI.git
cd ComplainceAI

2. Backend Setup

# Create virtual environment
cd backend
python -m venv venv

# Activate (Windows)
venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

3. Configure Environment Variables

Create backend/.env:

MONGODB_URL=mongodb+srv://<user>:<password>@<cluster>.mongodb.net/?appName=Cluster0
DATABASE_NAME=compliance_ai
SECRET_KEY=<your-secret-key>
ENCRYPTION_KEY=<your-encryption-key>
GOOGLE_API_KEY=<your-gemini-api-key>
GEMINI_MODEL_NAME=gemini/gemini-2.5-flash
GEMINI_EMBEDDING_MODEL=models/gemini-embedding-001
CELERY_BROKER_URL=mongodb+srv://<user>:<password>@<cluster>.mongodb.net/celery_db
CELERY_RESULT_BACKEND=mongodb+srv://<user>:<password>@<cluster>.mongodb.net/celery_db

Create secure-ai-dashboard/.env:

VITE_API_BASE_URL=http://localhost:8000

4. Start the Backend Server

cd ..
python -m uvicorn backend.main:app --reload --port 8000

5. Start the Celery Worker (separate terminal)

backend\venv\Scripts\celery -A backend.core.celery_app worker --loglevel=info -P solo

6. Frontend Setup (separate terminal)

cd secure-ai-dashboard
npm install
npm run dev

7. Open the Application

Navigate to http://localhost:5173 in your browser.

---

📸 Key Features

• Dashboard — Real-time KPIs, risk trends, violation charts
• AI Agents — 5 specialized agents + batch processing controls
• Violations — Flagged transactions with AI explanations
• Human Review — Resolve or Escalate with audit trail
• RAG Policies — Policy-grounded AI explanations
• Predictions — ML model output and confidence analysis
• Audit Logs — Complete activity history

---

👥 Team

• Mohita - Full Stack Developer & AI Engineer

• Abhay Tiwari - Full Stack Developer & AI Engineer

---

📄 License

This project was built for HackFest 2.0.