AI-Agent Document-Analyzer
Valentina Sanchez
Data Scientist
Data Analyst
AI Developer
LlamaIndex
Ollama
Python
This project is an AI-powered document analysis bot designed to process and extract information from PDF documents. It leverages the LlamaIndex library, LangChain, and Ollama models to parse, analyze, and interact with the content of these documents. The bot is capable of handling user queries and providing relevant information based on the PDFs it processes.
My goal with this project is two fold: to deepen my understanding of large language models (LLMs) and to employ Retrieval-Augmented Generation (RAG) to mitigate hallucinations that can occur with these models. By feeding the model my own curated information, I aim to create a tool that works effectively for my needs.
I wish I had access to something like this during my thesis work—having a tool that could compare and answer questions about the numerous research papers I had to read on computer vision would have been invaluable. My vision is to develop this into a tool that facilitates interactive conversations between the model and the user about any document they provide.
Keep in mind! This project is still a work in progress. My goal is to continuously enhance the bot by adding more tools to the agent, making it increasingly useful. Currently, one of the main areas I am working on is improving the clarity and readability of the output. By using a local Ollama model, I not only benefit from enhanced privacy, but also from cost savings, allowing me to experiment freely without worrying about additional expenses.
Features
PDF Parsing: Parses and interprets text from PDF documents using
LlamaParse.Vector Embedding: Converts document text into vector embeddings using a local model.
Vector Database: Stores vectorized document data in a vector-based database, allowing for efficient and accurate querying.
Query Engine: Utilizes the vector database to query and retrieve information from the documents.
ReAct Agent: Uses a ReActAgent to determine the appropriate tools and actions to answer user queries.
Interactive Bot: Engages in conversations, answering questions based on the content of the loaded documents.
Code Overview
LlamaParse: Used to parse PDF documents into a format that the program can process.
VectorStoreIndex: Converts document text into vectors and stores them in a vector-based database, allowing for efficient querying.
Query Engine: Wraps the vector database to provide a powerful querying interface.
ReActAgent: An agent that uses the llama3 model to handle reasoning and action decisions during interactions.
PromptTemplate: Defines the format for the bot’s responses.
