# 🧠 LLM-Based Insurance Policy Decision System

A document query system powered by large language models (LLMs) that extracts structured information from natural language insurance-related queries and returns clause-level decisions with justifications.

---

## 🚀 Features

- Parses age, gender, procedure, location, and policy duration from queries
- Uses semantic similarity to match insurance clauses
- Makes approval/rejection decisions based on rules
- Provides clause-level justifications
- Includes integration and performance testing

---

## 🗃️ Project Structure

| File/Folder                          | Description                              |
|-------------------------------------|------------------------------------------|
| `everything.py`                     | Main orchestrator                        |
| `parse.py`                          | Extracts structured data from query      |
| `test_parser.py`                    | Unit tests for parser                    |
| `test_decision_engine.py`          | Tests for decision logic                 |
| `test_integration.py`              | End-to-end system test                   |
| `test_performance.py`              | Performance evaluation                   |
| `enter_query.py`                   | Manual query entry for testing           |
| `all_clauses.json`                 | Raw clauses                              |
| `all_clauses_with_embeddings.json` | Clause embeddings                        |
| `clauses/`                          | Individual clause text files             |
| `models/`                           | Saved models and embeddings              |
| `output/`                           | Output and logs                          |
| `myenv/`                            | Virtual environment (ignored)            |

---

## 💻 How to Run

1. **Create a virtual environment**
   ```bash
   python -m venv myenv
   source myenv/bin/activate  # On Windows: myenv\Scripts\activate

2. Install dependencies

   pip install -r requirements.txt

3. Run the system

   python everything.py

4. Enter test query manually

   python enter_query.py

---

✅ Example Query

"A 60-year-old male wants a knee replacement in Bangalore, 8 months into his policy."

Output:

• Decision: Approved ❎ / Rejected ✅
• Clause: "Treatment covered after 6 months for age > 50"
• Explanation: "The user's policy duration satisfies the clause condition for coverage."

---

📌 Notes

• LLM used: sentence-transformers/all-MiniLM-L6-v2
• Clause embedding file is precomputed and saved in JSON format
• Performance tests benchmark query time and decision speed

---