README.md

---
license: apache-2.0
tags:
- decision-intelligence
- insurance
- governance
- audit
- gcc
library_name: bdr-decision-os
---

# BDR Decision Engine

## 🎯 Core Decision Operating System

**The foundational orchestration engine for insurance decision intelligence.**

This is NOT a standalone model. It is the **reusable core engine** that powers all insurance decision systems in the Bader AI platform.

---

## Architecture Principle

**Capabilities are NOT projects.**  
**Insurance decision systems ARE projects.**

This engine is industry-agnostic. Insurance logic is injected via configuration.

---

## What This Engine Does

The BDR Decision Engine provides:

### 1. Agent Orchestration
- Multi-agent workflow coordination
- Sequential and parallel execution
- Agent state management
- Error handling and recovery

### 2. Input Normalization
- Schema validation
- Data type conversion
- Missing value handling
- Format standardization

### 3. Decision Routing
- Rule-based routing
- Confidence-based escalation
- Priority queue management
- Load balancing

### 4. Governance Hooks
- Pre-decision validation
- Post-decision logging
- Compliance checks
- Regulatory alignment (IFRS, AML)

### 5. Metrics Hooks
- Performance tracking
- Confidence distribution
- Escalation rate monitoring
- Turnaround time measurement

### 6. Human-in-the-Loop Overrides
- Manual decision override
- Override justification capture
- Audit trail preservation
- Feedback loop integration

---

## Standard Decision Contract

All systems using this engine output the same contract:

```json
{
  "decision": "approve | reject | escalate | investigate | adjust",
  "confidence": 0.0-1.0,
  "rationale": "human-readable explanation",
  "risk_signals": ["fraud", "coverage_gap", "inconsistency"],
  "recommended_next_step": "string"
}
```

---

## Insurance Systems Using This Engine

This engine powers 4 production insurance systems:

| System | Purpose | Capabilities Activated |
|--------|---------|------------------------|
| **ClaimsGPT** | Claim approval decisions | vision_ocr, document_parser, vision_vqa, financial_analysis, risk_scoring, decision_optimizer |
| **FraudSimulator-AI** | Fraud detection | anomaly_detection, scenario_simulator, risk_scoring, financial_analysis |
| **AutoRiskScoreEngine** | Underwriting & pricing | financial_analysis, optimization_engine, time_series_forecasting, scenario_simulator |
| **InsuranceKnowledgeAgent** | Policy interpretation | language_rag, policy_clause_reasoner, document_parser |

---

## Capability Module Interface

Each capability module must expose:

```python
{
  "input_schema": {},
  "output_schema": {},
  "supported_decisions": []
}
```

### Available Capability Categories

**Vision & Documents (3 modules)**
- `vision_ocr` - Extract text from images
- `vision_vqa` - Visual question answering
- `document_parser` - Parse structured documents

**Language & Knowledge (3 modules)**
- `language_rag` - Retrieval-augmented generation
- `text_analysis` - Text classification and NER
- `policy_clause_reasoner` - Policy interpretation

**Risk & Analytics (4 modules)**
- `anomaly_detection` - Detect unusual patterns
- `risk_scoring` - Calculate risk scores
- `financial_analysis` - Financial impact assessment
- `time_series_forecasting` - Predict trends

**Decision Intelligence (2 modules)**
- `decision_optimizer` - Optimize decision outcomes
- `scenario_simulator` - Simulate decision scenarios

**Governance (4 modules)**
- `audit_logger` - Log all decisions
- `confidence_tracker` - Track confidence scores
- `override_handler` - Manage human overrides
- `drift_monitor` - Detect model drift

---

## Governance Guarantees

✅ **Every decision is logged**  
✅ **Rationale is always stored**  
✅ **Confidence is tracked**  
✅ **Human overrides are recorded**  
✅ **Audit replay is supported**  

Governance is **shared across all systems** - never duplicated.

---

## Usage Pattern

```python
from bdr_decision_engine import DecisionEngine

# Initialize engine with config
engine = DecisionEngine(
    system="claims",
    enabled_capabilities=[
        "vision_ocr",
        "document_parser",
        "risk_scoring",
        "decision_optimizer"
    ],
    governance="strict",
    metrics="claims_kpis"
)

# Process decision
result = engine.process({
    "claim_id": "CLM-2026-001",
    "claim_type": "motor",
    "claim_amount": 5000,
    "documents": [...],
    "images": [...]
})

print(result.decision)      # "approve"
print(result.confidence)    # 0.92
print(result.rationale)     # "Claim meets all criteria..."
print(result.risk_signals)  # []
```

---

## Strategic Positioning

**Bader AI — Insurance Decision Operating System**

This engine is the foundation of the GCC reference architecture for insurance AI, designed for:

- **Insurers** - Production decision automation
- **Regulators** - Auditable AI governance
- **System Integrators** - Reusable decision infrastructure

---

## What This Prevents

❌ Demo chaos  
❌ Capability sprawl  
❌ Architectural drift  
❌ Ungoverned AI outputs  
❌ Non-insurance use cases  

---

## License

Apache 2.0

---

## Organization

[BDR AI Organization](https://huggingface.co/bdr-ai-org)

**Building the future of insurance decision intelligence for the GCC region.**