This project contains a few scripts for understanding COBOL code that I did for an interview. It uses instructor and Pydantic to create a report about COBOL code. It embeds the GNU COBOL 2.0 Programmers manual using llama-index and presents a chatbot-like interface for RAG.

It also contains a script json_to_markdown.py for turning the JSON output from instructor/GPT into a report.

The vector store is kept offline once you create the embeddings. I think you can just use my embeddings from this repo though and avoid the 20 cent charge.

poetry install
poetry shell

python ./cobal_reporter/rag.py

First, add any additional COBOL source code documents to the test_queries list in openpipe_tuner.py.

python ./cobol_reporter/openpipe_tuner.py

Then, turn the JSON to Markdown:

python ./cobol_reporter/json_to_markdown.py

Happy trails prospector 😉