pdf chatbot
My Projects

PDF Chatbot

Project Report: PDF Chatbot with Mistralai
Project Duration: 2024/09/24 – 2024/09/29 (12 hours)

Project Overview:

The objective of PDF Chatbot project was to develop an Artificial Intelligence chatbot capable of interacting with the content of a book provided in PDF format. Users can ask questions or discuss the content directly with the chatbot. The chatbot leverages advanced language models and embeddings to offer accurate and context-aware responses based on the book’s content.

Technical Approach for PDF Chatbot:

  1. Model Selection:
    We opted for Mistralai, an open-source large language model (LLM), for its ease of use and availability compared to proprietary models like GPT-4 and Meta’s LLAMA. Mistralai allowed us to bypass access restrictions and still delivered robust performance for our needs.
  2. Text Preprocessing:
    The project involved receiving a PDF file as input, which was then split into manageable chunks. This step ensured that the LLM could efficiently handle the text, facilitating accurate question-answer interactions.
  3. Embedding and Vectorization:
    Using Hugging Face Embeddings, we transformed the text chunks into vector representations. These embeddings were stored and managed using chromedb, which enabled efficient searching and retrieval of relevant text portions during interactions.
  4. Retrieval-Augmented Generation (RAG):
    We employed the RAG approach, where the chatbot searches for the most relevant chunks from the book and then generates answers using the LLM. This combination ensures that the responses are both accurate and contextually grounded in the content of the PDF.
  5. Fine-Tuning (Optional):
    The system is designed to accommodate further fine-tuning of the LLM using additional tokens and enhanced embedding strategies, improving the overall interaction quality for more complex queries.

Results for PDF Chatbot:

The chatbot successfully processes PDF books, allowing users to ask questions and receive relevant, content-driven answers. The integration of Mistralai with RAG and fine-tuning mechanisms ensures efficient and accurate responses.

Repository:

For further details and implementation specifics, the code can be found at my repository on GitHub:
PDF Chatbot