BikeShield: AI Claims Adjudicator

BikeShield is a Multi-Agent AI System that processes bike repair invoices, audits them against live market prices, and determines the correct settlement amount according to insurance policy rules.

Agents Overview

1. Ingestion Agent: Checks image quality (resolution, brightness).
2. Parsing Agent: Extracts line items from the bill using Gemini 2.0.
3. Anomaly Detection Agent: Performs live Google Search to verify part prices in India.
4. Policy Agent: Calculates depreciation (Plastic 50%, Metal 10%, etc.).
5. Investigator Agent: Synthesizes a final "Risk Story" and verdict.

Setup

1. Install Dependencies:

   pip install -r requirements.txt

2. Environment Variables: Ensure you have a .env file in this directory with your Google API Key:

   GOOGLE_API_KEY=your_actual_api_key_here
   

How to Run

To process a claim bill, run the main.py script with the path to the bill image:

python main.py "path/to/your/bill_image.jpg"

Example

python main.py "Sample Images/Gemini_Generated_Image_a2dl79a2dl79a2dl.png"

Output

The pipeline generates the following outputs:

• Console Output: Live logs of each agent's progress.
• final_case_report.md: The final investigation report with the verdict (Approve/Reject) and financial breakdown.
• full_pipeline_output.json: A detailed JSON file containing the raw output from every agent (parsing data, audit results, specific depreciation calculations).
• bikeshield_pipeline.log: Debug logs for troubleshooting.