AI-powered DDR Generator

This project provides a minimal backend system for generating Detailed Diagnostic Reports (DDR) by extracting information from two PDF documents (Inspection Report and Thermal Report) and leveraging an LLM to structure the summary.

Tech Stack

• Python (>=3.8)
• FastAPI
• PyMuPDF (fitz)
• OpenAI (GPT) or any LLM via environment key

Folder Structure

ddr-ai-system/
│
├── main.py
├── pdf_parser.py
├── image_extractor.py
├── ai_processor.py
├── report_generator.py
├── requirements.txt
├── README.md
└── uploads/              # stores uploaded PDFs and extracted images

Setup Instructions

1. Clone or copy this repository to your local machine.

2. Create a virtual environment and activate it:

   python -m venv venv
   source venv/bin/activate    # macOS/Linux
   venv\Scripts\activate     # Windows

3. Install dependencies:

   pip install -r requirements.txt

4. Set your LLM API key.

   • Option A (recommended): create a .env file in the project root.

     Example .env:

     GEMINI_API_KEY=your_key_here
     AI_PROVIDER=gemini

   • Option B (environment variables):

     For OpenAI:

     export OPENAI_API_KEY="your_key_here"         # macOS/Linux
     setx OPENAI_API_KEY "your_key_here"          # Windows (requires new shell)

     For Gemini:

     export AI_PROVIDER=gemini
     export GEMINI_API_KEY="your_key_here"

   Note: the server must be started after setting env vars so they are visible to the running process.

5. Run the server:

   uvicorn main:app --reload

6. Use the endpoint:

   • POST /generate-ddr with form-data fields inspection_report and thermal_report both as file uploads.
   • Returns JSON containing extracted text, image paths, and the generated DDR report.

Example cURL

curl -X POST "http://127.0.0.1:8000/generate-ddr" \
  -H "accept: application/json" \
  -H "Content-Type: multipart/form-data" \
  -F "inspection_report=@inspection.pdf;type=application/pdf" \
  -F "thermal_report=@thermal.pdf;type=application/pdf"

Simple Streamlit UI

You can also launch a basic upload form and display results with Streamlit.

streamlit run ui.py --server.maxUploadSize=200

The interface allows you to choose two PDFs and shows the DDR output inline. (Note: Streamlit upload size is controlled via command line/config, not in Python code.)

Notes

• Image extraction uses PyMuPDF; if the PDFs have no images, the images list will be empty.
• The LLM prompt enforces structure and rules; if the model returns non-JSON, the raw output is returned under the raw key.
• Uploaded files and extracted images are stored in the uploads/ directory.

Feel free to extend functionality, add authentication, or swap in a different PDF parser or LLM provider.