Transforming Management Reporting with Generative AI (LLM)

1. Objective

This project is a proof-of-concept demonstrating how to combine data analysis techniques with Large Language Models (LLMs) to automatically generate management reports from raw financial data stored in Excel format.

The goal is to streamline the report generation process, providing data-driven insights with AI-generated summaries and visualizations, a life changer for accounting and finance professionals. Watch demonstration video here.

2. Setup and Installation

Prerequisites

• Python+
• pipx package manager
• Ollama
• open-webui

Installation Steps

1. Clone the Repository:

   git clone https://github.com/voidbydefault/LLM_Financial_Report_Generator.git
   cd LLM_Financial_Report_Generator

2. Create and Activate a Virtual Environment (Recommended):

   python3 -m venv .venv
   source .venv/bin/activate  # On Linux/macOS
   .venv\Scripts\activate   # On Windows

3. Install Dependencies:

   pip install -r requirements.txt

3. Install Ollama

This project uses a local LLM server for generating text commentary. You will need to install Ollama.

1. Download and install Ollama: Follow the instructions on the Ollama website for your operating system.

2. Verify Installation: After installing, run ollama --version in your terminal to verify that Ollama is installed correctly.

4. Install Open WebUI (Optional, for a User Interface)

To interact with your Ollama models via API for this project (or through a browser-based interface in General), you can install Open WebUI:

• Using pipx (Recommended):

  pipx install open-webui

  After installation, you can run with

  open-webui

• Install from source Refer to https://github.com/open-webui/open-webui for more details.

5. Running the Code

1. Prepare Your Data: This project comes with a demo Excel file named 1k_lines_sales_data.xlsx.

2. Configure Model (Optional):

   • In main.py, you can select the desired LLM model by uncommenting the corresponding line:

   model_name = "phi4:latest"
   # model_name = "gemma3:12b"
   # model_name = "deepseek-r1:1.5b"

3. Run the Main Script:

   python main.py

The script will generate an analysis results Excel file, visualizations, a markdown report, and finally a Word document in the output directory.

Once you are acquainted with the process, bring your data instead of 1k_lines_sales_data.xlsx and update file name and column headers (use ChatGPT or Deekseek to do this).

6. Hardware Requirements

The hardware requirements depend heavily on the chosen LLM model (larger models require more GPU VRAM).

• Model Options (Adjust model_name in main.py):

  • phi4:latest: Offers best results; requires approximately 16 GB of GPU RAM.

    model_name = "phi4:latest" # gives best results, takes 16 GB of GPU RAM

  • gemma3:12b: Offers 2nd best results; requires approximately 12 GB of GPU RAM.

    model_name = "gemma3:12b" # gives 2nd best results, takes 12 GB of GPU RAM

  • deepseek-r1:1.5b: Smallest model; requires approximately 3 GB of GPU RAM (May add <THINKING> or other nonesense at the end of response.).

    model_name = "deepseek-r1:1.5b" # smallest model takes 3 GB of GPU RAM adds <THINKING>

  Start with a smaller model if you have limited VRAM, and gradually increase the model size until you find a balance between performance and resource utilization.