FinReportAI - Infinity AI

App Description

FinReportAI is an on-device AI-powered financial report analysis tool designed for non-financial professionals. It extracts key data from lengthy financial reports using OCR, processes and visualizes the data, and generates an easy-to-understand financial analysis report using NLP and a local language model. This enables users to receive personalized financial insights and investment advice without needing extensive financial expertise.

Developers

• Anqi Guan

  • LinkedIn: https://www.linkedin.com/in/anqi-g-a2108b24b/
  • Email: [email protected]

• Zhendong Yang

  • LinkedIn: https://www.linkedin.com/in/zhendong-aaron-yang-98667b290/
  • Email: [email protected]

• Jing Du

  • LinkedIn: https://www.linkedin.com/in/du-jing2-northeastern-edu/
  • Email: [email protected]

• Sicheng Zhou

  • LinkedIn: https://www.linkedin.com/in/sicheng-zhou-81419758/
  • Email: [email protected]

• Rui Min

  • LinkedIn: https://www.linkedin.com/in/ryanmin55
  • Email: [email protected]

Features

• PDF to Image Conversion:
  Uses pdf2image and PyMuPDF (fitz) to convert PDF pages to high-resolution PNG images.

• OCR Processing:
  Uses Tesseract OCR (via pytesseract and Pillow) to extract text from images.

• Data Extraction & Parsing:
  Processes the OCR output to detect date columns, extract financial field values, and capture company information.

• Data Analysis & Visualization:
  Uses pandas, matplotlib, and seaborn to analyze financial data and generate visualizations.

• Report Generation:
  Utilizes a language model (via llama-cpp-python or Hugging Face's transformers pipeline) to generate a comprehensive financial report.

• Web UI:
  A simple Flask-based file upload interface for users to upload PDFs and trigger processing.

Installation

Prerequisites

• Python 3.8 or later
• Tesseract OCR installed and added to your system's PATH
  • Windows: Download Tesseract
  • macOS: brew install tesseract
  • Linux: sudo apt-get install tesseract-ocr

Setup Instructions

To set up the project from scratch, follow these steps:

1. Clone the Repository:

   git clone https://github.com/YourGitHubUsername/FinReportAI.git
   cd FinReportAI
   

2. Create a Virtual Environment (Recommended):

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

3. Install Dependencies:

   pip install -r requirementss
   

   Note: The requirements.txt file includes dependencies such as pdf2image, numpy, pillow, matplotlib, seaborn, pytesseract, pandas, flask, transformers, torch, and the appropriate LLM package (e.g., llama-cpp-python).

Or Install the required Python packages by running:

```bash
pip install pdf2image numpy pillow matplotlib llama-cpp-python pytesseract pandas flask seaborn transformers torch subprocess markdown2
```

Note: If you encounter issues installing llama_cpp or if you need a version that doesn't require AVX, consider using llama-cpp-python instead (see its GitHub repository for more details).

4. Set Up Tesseract:

   • Windows: Download and install Tesseract OCR from Tesseract at UB Mannheim and add it to your PATH.
   • macOS: Install via Homebrew:

     brew install tesseract
     

   • Linux: Install via package manager:

     sudo apt-get install tesseract-ocr
     

5. Configure the Application:

   • Place your financial report PDF in the src/uploads folder.
   • Ensure that the outputs folder exists (the application will create it automatically if it doesn't).

Project Structure

your_project/
├── src/                     
│   ├── app.py                   # Flask file upload UI and main application
│   ├── advice_pipeline.py                   # Generate a financial advice report
│   ├── ocr_pipeline.py             # PDF-to-image conversion, OCR extraction, and parsing OCR text into financial_data.csv
│   ├── visualization_pipeline.py # Visualization and analysis of financial data (e.g., plotting trends)
│   ├── report_pipeline.py       # Generate a comprehensive financial analysis report using an LLM
│   ├── outputs/                 # Automatically generated output files:
│   │   ├── extracted_text.txt   # OCR results
│   │   ├── financial_data.csv   # Parsed financial data CSV
│   │   ├── analysis_report.txt  # Generated financial analysis report
|   |   └── advice_report.txt    # Generated financial advice report
│   └── uploads/                 # Directory for uploaded PDF files
├── examples/                    # Example scripts or notebooks for testing/experiments
└── README.md                    # This file

Usage

1. Run the Web UI for File Upload

1. Open a terminal and navigate to the source folder:

   cd src
   

2. Run the Flask application:

   python app.py
   

3. In your browser, go to http://127.0.0.1:5000/ and upload your financial_report.pdf file from the examples folder.

2. Convert PDF to Images & Parse OCR Data into CSV

To run the OCR pipeline:

1. Place your financial_report.pdf in the uploads folder.
2. From the src folder, run:

   python ocr_pipeline.py
   

   This script will:
   • Convert the PDF from the uploads/financial_report.pdf into PNG images stored in the outputs/ folder.
   • Apply Tesseract OCR on each image and save the combined text to outputs/extracted_text.txt.
   • Read outputs/extracted_text.txt, detect date columns and extract financial fields (and company name), then write the parsed data into outputs/financial_data.csv.

3. Generate the Financial Analysis and Advice Report

To generate a report based on the parsed CSV:

1. From the src folder, run:

   python report_pipeline.py
   python advice_pipeline.py
   python visualization_pipeline.py
   

2. The script will load outputs/financial_data.csv, construct a prompt, and use the offline language model to generate a detailed financial analysis report.
3. The final report is saved as outputs/analysis_report.txt and outputs/advice_report.txt.

Customization

• OCR Preprocessing:
  You can adjust the image preprocessing in the OCR script to improve recognition accuracy.

• Parsing Logic:
  Modify the regex patterns in the parsing scripts if your OCR output format changes.

• Model & Prompt:
  Adjust the prompt for the language model to fit your desired report style. If using a different LLM, update the model loading code accordingly.

Troubleshooting

• Tesseract Not Found:
  Make sure Tesseract is installed and its executable is in your PATH. You may specify the path in your script:

  pytesseract.pytesseract.tesseract_cmd = r"C:\Program Files\Tesseract-OCR\tesseract.exe"
  

• Dependency Issues:
  Ensure all dependencies are installed in your virtual environment. Use pip freeze to verify package versions.

• Model Compatibility:
  If you face issues with the language model (e.g., AVX requirements on Snapdragon devices), consider using an ARM-compatible version like llama-cpp-python.

Open-Source License

This project is licensed under the MIT License.

For help choosing an open-source license, visit choosealicense.com.

Acknowledgments

• PyMuPDF (fitz)
• Tesseract OCR
• llama-cpp-python
• Transformers
• Flask