DepoIndex

DepoIndex is an automated, AI-powered workflow that reads deposition transcripts, detects every distinct subject discussed, and produces a Table of Contents (TOC). For each topic, the system lists the starting page number and line number, then outputs the results as a chronologically ordered Topic Index.

DepoIndex enables clerks to generate an accurate, paginated topic index from a deposition in minutes - eliminating manual scanning and ensuring that judges can jump directly to any point of interest.

To-Do

• Topic Extraction
• Table of Contents Generation
• Validation Notebook
• CLI wrapper

Requirements

• Python 3.12
• Ollama with the following models pulled:
  • Gemma3 (4B) for topic extraction.
  • DeepSeek-R1 (8B) for reasoning based validation.

Setup

1. Clone repo:

   git clone https://crowaltz24/DepoIndex
   cd DepoIndex

2. Ensure your directory structure is as follows:

   DepoIndex
   |_ /inputs
   |_ /outputs
   |_ /utils
   |_ /validation
   

3. Create venv:

   python -m venv venv
   venv/scripts/activate

4. Install requirements:

   pip install -r requirements.txt

How to Use

CLI

1. Place your Deposition Transcript in /inputs.

2. Run the script, providing input and output file names.

   python build_toc.py --file deposition.pdf --out toc.docx

• --file accepts input path (relative to /inputs)
  • This may be in .pdf or .txt format.
• --out accepts output path (relative to /outputs)
  • This may be in .md or .docx format

3. Access your outputs from the /outputs directory.

Manual Usage

Run depoindex.py.

OR

Run the scripts one by one:

• topic_extraction.py extracts topics into extracted_topics.json.
• topc_generator.py uses the extracted topics to generate a table of contents, saving it in Markdown and docx formats.

Validation

validation.ipynb is... a validation notebook.

It takes a random sample of topics from y our extracted topics, and runs them past a reasoning LLM to compare with an excerpt of the text they were generated from to judge accuracy.

NOTE: You can change the number_of_topics to validate on as many topics as you want.

License

MIT License.