auto_report_gen

auto_report_gen is a tool designed to streamline and partially automate the redaction of annual reports. It assists in reusing last year’s report structure, integrating updated project information, and generating high-quality section drafts using GPT-based models.

---

✨ Overview

Writing annual reports each year often involves reusing the same structure, updating project contributions, and ensuring consistency in tone and length. auto_report_gen automates most of this workflow by:

• Parsing last year’s report into structured sections using Docling.
• Allowing you to map new project reports (PDFs) to these sections via a Streamlit interface.
• Automatically generating updated text through a customized prompt pipeline powered by the ChatGPT API (model: GPT-5 Mini).
• Running all section generations concurrently for speed.

---

🚀 Features

1. Interactive Frontend (Streamlit)

A user-friendly web interface guides you through the entire process, from uploading files to downloading the final report.

2. Upload Last Year’s Report

• Upload your previous year's PDF.
• The system automatically parses it into major sections using Docling's advanced document understanding.

3. Upload & Link Resources

• Upload multiple new resource PDFs (project reports, updates, etc.).
• Interactively link specific resources to specific sections of the report.
• Select/Unselect All functionality for quick configuration.

4. Custom Instructions

• Add manual comments or instructions for specific sections (e.g., "Emphasize the new partnership", "Keep this brief").
• These comments are directly injected into the generation prompt.

5. Smart Generation Pipeline

For every section, a tailored prompt is assembled containing:

• The text from last year's corresponding section (for style and context).
• The text from the linked new resources.
• Your manual instructions.
• Internal guidelines ensuring consistency in tone, format, and length.

6. Parallel Execution

• All section prompts are sent to the ChatGPT API in parallel.
• Uses GPT-5 Mini (or configured model) for high-quality text generation.

---

🛠️ Requirements

• Python ≥ 3.13
• Access to the OpenAI/ChatGPT API (API Key required)
• uv (recommended) or pip for dependency management

---

📦 Installation

This project uses uv for fast dependency management, but can also be installed via standard pip.

Option 1: Using uv (Recommended)

1. Install uv (if not already installed):

   pip install uv
   

2. Clone the repository:

   git clone https://github.com/your-username/auto_report_gen.git
   cd auto_report_gen
   

3. Sync dependencies:

   uv sync
   

Option 2: Using pip

1. Clone the repository:

   git clone https://github.com/your-username/auto_report_gen.git
   cd auto_report_gen
   

2. Install dependencies:

   pip install .
   

---

▶️ Usage

1. Set up your API Key: Create a .env file in the root directory and add your OpenAI API key:

   OPENAI_API_KEY=sk-...
   

2. Run the Streamlit App:

   # If using uv
   uv run streamlit run front_end/main_frontend.py
   
   # If using standard pip/venv
   streamlit run front_end/main_frontend.py
   

3. Workflow:

   • Step 1: Upload Last Year's PDF and click "Parse Report".
   • Step 2: Upload New Resource PDFs.
   • Step 3: Expand sections to:
     • Enable/Disable them.
     • Link specific resource files.
     • Add manual comments.
   • Step 4: Click "Generate Report".
   • Step 5: Download the final Markdown report.

---

🤖 Model & API

• Models: Defaults to gpt-5-mini and gpt-5-nano (ensure your API key has access to these models).
• Docling: Used for robust PDF parsing and chunking.

---

📄 License

TBD

---

🙌 Contributions

Pull requests and suggestions are welcome! Feel free to open an issue to report bugs or request features.