This is a small Python utility that empowers users to read, summarize, and ask questions about PDF documents using Open AI Apis.
Here is a short video demonstrating loading, batch_summarizing, vectorizing, and asking questions about a PDF document.
To use this tool you must have an Open AI Api key, as that is the model used for both text generation and embeddings.
The environment variable OPENAI_API_KEY must be set to your API key for this tool to work.
- Clone down the repository locally.
- (Optional, but recommended) Setup a virtual python environment.
virtualenv venv && source venv/bin/activate - Install the Python requirements with
pip install -r requirements.txt. - Drop a PDF you want to read in the same directory. (Can be anywhere on your filesystem, but local is easier)
- Run the tool
python doc_assist.py
After these steps you will have a prompt called doc assist: you can use to interact with your documents.
doc assist: load <path_to_pdf_file>
This loads the PDF file into memory and is the required first step.
doc assist: read <page_num>
This reads the text of a given page of the PDF document, and is useful to quickly inspect pages on the fly.
doc assist: summarize <page_num>
Generates a short summary for a specific page.
doc assist: batch_summarize <start_page> <end_page>
start_page and end_page are optional. If not specified the entire document is used.
This uses a MapReduce system to iteratively generate summaries of each page, and then generate holistic summaries from there.
doc assist: vectorize
This creates a ChromaDB collection for the PDF document, creates embeddings for the document pages, and persists the content for later use.
doc assist: ask <query>
Ask questions about the content of the PDF document. vectorize must be run first.