Project Baitak: Sovereign Banking Intelligence Suite

Codename: Project Baitak (صانع الذكاء المالي)
Status: Alpha Release
Version: 1.0.0
License: MIT

Vision: Sovereign Intelligence for the GCC Banking Sector

Project Baitak is a sophisticated, self-learning banking intelligence platform designed to unlock actionable insights from the GCC's largest Islamic financial institutions. Using publicly available financial data from Kuwait Finance House (KFH) and other regional banks, Baitak synthesizes 12 specialized machine learning models into a unified Orchestrator Agent that generates dual-language (English/Arabic) strategic insights.

The platform is built for financial analysts, risk managers, and institutional investors who need real-time, AI-driven intelligence on Islamic banking performance, regulatory compliance, and market dynamics in the GCC.

Core Architecture

1. 12 Specialized Intelligence Models

Baitak orchestrates 12 AutoML models across four domains:

Retail Intelligence (3 models)

Churn Prediction: Identifies high-risk customer segments using transaction patterns and engagement metrics.
Customer Lifetime Value (CLV): Forecasts long-term customer profitability based on tenure, revenue, and financing exposure.
Product Propensity: Predicts likelihood of customer adoption for Murabaha (cost-plus financing) and Wakala (agency) products.

Risk Intelligence (3 models)

Credit Scoring (Islamic Finance Compliant): Assesses default risk while adhering to Shariah principles (no interest-based penalties).
Fraud Detection (Anomaly): Identifies suspicious transaction patterns using Isolation Forest algorithms.
CBK 2026 Regulatory Stress-Testing: Simulates capital adequacy ratios under adverse market scenarios per Central Bank of Kuwait requirements.

Operational Intelligence (2 models)

Demand Forecasting (Branch Liquidity): Predicts daily liquidity needs across branch networks based on customer footfall and seasonality.
Sales Anomaly Detection: Flags unusual sales volumes or product mix shifts that may indicate operational issues or fraud.

Market Intelligence (2 models)

Sukuk Yield Prediction: Forecasts Islamic bond yields based on oil prices, market indices, and global interest rates.
Real Estate Portfolio Risk: Assesses risk exposure for KFH's significant real estate financing portfolio.

Institutional Bridge (1 model)

AgenThink Institutional Stub: Placeholder for future three-tier ecosystem (Personal → Business → Institutional).

2. Orchestrator Agent

The Meta-Orchestrator is a Python-based intelligent router that:

Analyzes incoming queries to determine which models to invoke.
Retrieves historical insights from a ChromaDB vector database for context.
Synthesizes model outputs with LLM-generated analysis.
Generates dual-language insights (English & Arabic) using Gemini API.

User Query
    ↓
Meta-Orchestrator (Intent Classification)
    ↓
Model Selection (Retail/Risk/Ops/Market)
    ↓
ChromaDB Memory Retrieval
    ↓
Model Inference
    ↓
Gemini LLM (Dual-Language Synthesis)
    ↓
Output: English + Arabic Insight

3. Memory Layer

ChromaDB stores quarterly insights and KPIs, enabling the Orchestrator to:

Recall historical trends ("What was NFM in Q3 2025?").
Identify anomalies ("Cost-to-Income improved by 140 bps—why?").
Support multi-turn conversations with persistent context.

Key Financial KPIs Tracked

Based on KFH 2024/2025 Annual Reports:

KPI	FY 2024	FY 2025	Y-o-Y Change
Net Profit (Shareholders)	KD 601.8M	KD 632.1M	+5.0%
Net Financing Income	KD 1,147.0M	KD 1,279.2M	+11.5%
Cost-to-Income Ratio	35.46%	34.06%	-140 bps ✓
Total Assets	KD 36.7B	~KD 38.5B	+5%
Capital Adequacy Ratio	19.89%	~20%	Stable
EPS (Adjusted)	33.68 Fils	35.64 Fils	+5.8%

Installation & Setup

Prerequisites

Python 3.9+
pip or uv package manager
Gemini API key (for LLM insights)

Quick Start

# Clone the repository
git clone https://github.com/yourusername/Project-Baitak-Intelligence.git
cd Project-Baitak-Intelligence

# Install dependencies
pip install -r requirements.txt

# Configure environment
cp .env.example .env
# Edit .env and add your GEMINI_API_KEY

# Train models (one-time)
python3 models/retail/retail_models.py
python3 models/risk/risk_models.py
python3 models/operations/operations_models.py
python3 models/market/market_models.py

# Run the Orchestrator Agent
export GEMINI_API_KEY="your_key_here"
python3 src/orchestrator/orchestrator_agent.py

Example Queries

from src.orchestrator.orchestrator_agent import OrchestratorAgent

agent = OrchestratorAgent()

# Query 1: NFM Trend Analysis
response = agent.process_query("What is the current NFM trend and how does it relate to digital transformation?")
print(f"English: {response['english_insight']}")
print(f"Arabic: {response['arabic_insight']}")

# Query 2: Churn Risk
response = agent.process_query("Predict customer churn for a high-balance customer.")
print(f"English: {response['english_insight']}")

# Query 3: Regulatory Compliance
response = agent.process_query("Summarize the latest insights on KFH's cost-to-income ratio.")
print(f"English: {response['english_insight']}")

Project Structure

Project-Baitak-Intelligence/
├── models/
│   ├── retail/
│   │   ├── retail_models.py          # Churn, CLV, Propensity
│   │   └── retail_suite.pkl          # Trained models
│   ├── risk/
│   │   ├── risk_models.py            # Credit, Fraud, Stress-Testing
│   │   └── risk_suite.pkl
│   ├── operations/
│   │   ├── operations_models.py       # Liquidity, Sales Anomaly
│   │   └── operations_suite.pkl
│   └── market/
│       ├── market_models.py           # Sukuk, Real Estate
│       └── market_suite.pkl
├── src/
│   ├── memory/
│   │   └── chroma_memory.py           # ChromaDB vector store
│   ├── orchestrator/
│   │   └── orchestrator_agent.py      # Meta-Orchestrator + LLM
│   └── __init__.py
├── data/
│   └── chroma_db/                     # ChromaDB persistent storage
├── docs/
│   ├── kfh_research_summary.md        # KFH 2024/2025 analysis
│   └── architecture.md                # Detailed architecture
├── requirements.txt                   # Python dependencies
├── .env.example                       # Environment template
├── README.md                          # This file
└── LICENSE                            # MIT License

Dependencies

See requirements.txt for the full list. Key packages:

scikit-learn==1.8.0          # AutoML models
pandas==2.0+                 # Data manipulation
numpy==1.24+                 # Numerical computing
chromadb==0.3+               # Vector database
google-generativeai==0.3+    # Gemini API

Configuration

Environment Variables

Create a .env file in the project root:

# Gemini API Configuration
GEMINI_API_KEY=your_gemini_api_key_here

# Optional: Data paths
DATA_PATH=/home/ubuntu/Project-Baitak-Intelligence/data
MODELS_PATH=/home/ubuntu/Project-Baitak-Intelligence/models

# Optional: Logging
LOG_LEVEL=INFO

Usage Examples

Example 1: Retrieve Historical Insights

from src.memory.chroma_memory import BaitakMemory

memory = BaitakMemory()

# Query for NFM insights
results = memory.query_insights("What was the Net Financing Margin in 2025?", n_results=5)
for doc, meta in zip(results["documents"], results["metadatas"]):
    print(f"Insight: {doc}, Metadata: {meta}")

Example 2: Run Model Predictions

from models.retail.retail_models import RetailIntelligence

ri = RetailIntelligence()
ri.train_models()  # Train on synthetic data

# Predict churn for a customer
synthetic_customer = {
    'avg_balance': 6000,
    'transaction_count': 20,
    'digital_usage_score': 0.8,
    'murabaha_active': 1
}
churn_prediction = ri.models['churn_prediction'].predict([list(synthetic_customer.values())])
print(f"Churn Risk: {churn_prediction[0]}")

Example 3: Generate Dual-Language Insights

from src.orchestrator.orchestrator_agent import OrchestratorAgent

agent = OrchestratorAgent()
response = agent.process_query("Tell me about KFH's digital transformation initiatives.")
print(f"🇬🇧 English: {response['english_insight']}")
print(f"🇸🇦 Arabic: {response['arabic_insight']}")

Roadmap

Phase 1 (Current)

✅ 12 AutoML models trained on synthetic data
✅ Orchestrator Agent with ChromaDB memory
✅ Dual-language insight generation

Phase 2 (Planned)

Real-time data ingestion from KFH API (if available)
Advanced NLP for intent classification
Dashboard for visualization
REST API for third-party integration

Phase 3 (Future)

Integration with AgenThink Institutional tier
Multi-bank support (Ahli United, NBK, etc.)
Regulatory compliance reporting
Predictive alerts for risk events

Data Sources

Primary Research

KFH 2024 Annual Report: Comprehensive financial statements, governance, and strategy.
KFH FY-2025 Earnings Presentation: Latest quarterly results (Feb 2026).
Boursa Kuwait: Market data, trading volumes, and indices.
Central Bank of Kuwait (CBK): Regulatory guidelines and monetary policy.

Synthetic Data

All 12 models are trained on synthetically generated data to ensure:

Privacy compliance (no real customer data)
Reproducibility for testing
Scalability for different scenarios

Contributing

Contributions are welcome! Please follow these guidelines:

Fork the repository.
Create a feature branch (git checkout -b feature/your-feature).
Commit your changes (git commit -m 'Add your feature').
Push to the branch (git push origin feature/your-feature).
Open a Pull Request.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Disclaimer

Project Baitak is for educational and research purposes only. The insights generated are based on publicly available data and synthetic models. They should not be used as the sole basis for investment or financial decisions. Always consult with qualified financial advisors before making any investment decisions.

Contact & Support

For questions, issues, or suggestions, please open an issue on GitHub or contact the development team.

Project Baitak — Sovereign Intelligence for the GCC Banking Sector 🦅📊

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support