Use of Document for Question Answering Task

[aaroosh-07]

Description

As user cannot provide context through text box each time, we need to extend our functionality to different document formats such as .pdf, .docx etc. We will add support for pdf files first and then support for other file formats.

Discussion needed

• [ ] How to extract text from pdf file
• [ ] How to prompt LLM models for large context

  Acceptance Criteria

• [ ] modify the experimental script to take pdf files as context and perform question answering task on it.

[aaroosh-07]

I looked into how we can use documents for question answering task. We will need to implement a RAG, which will convert the documents into chunks. Then these chunks will be converted into vector embedding and stored into vector database, and later we can use the query to retrieve chunks from vector DB and use these chunks to perform question answering task. One thing we can look into is, Can we skip using vector DB and store chunks in some file and then fetch chunks by implementing cosine similarity.

[aaroosh-07]

We will start working on this by creating a script in experimentation folder. and after successful implementation add this feature to web app