Multi-Agent-Retrieval-Augmented-Generation-System

The System is designed to query and retrieve relevant information from a document set using multiple specialized Agents.

Note: Due to resource constraints the "facebook/bart-large-cnn" was used for final output generation which affects the generation quality.

Incase GPU is available the model can be replaced (in generation.py) to any other hugging face model like"google/bigbird-pegasus-large-arxiv".For most appropriate results the Gpt-4 model can be used through Openai API.

System Flow showing interaction between multiple agents.

System Overview

• Agent 1: Query Parser
  • Analyzes user queries to extract key terms or topics.
• Agent 2: Document Retriever
  • Searches the document database for relevant files.
• Agent 3: Document Ranker
  • Ranks the retrieved documents by relevance.
• Agent 4: Response Generator
  • Synthesizes a concise answer from the top-ranked documents.

Files Overview

• app.py

  • The main file that serves as the entry point to run the system. It orchestrates the flow of the program by invoking the other modules in the correct sequence.

• process.py -To preprocess .pdf files ,extract text , Preprocess (tokenization and cleaning),Generate Embeddings using Sentence-Transformer, Store embeddings in document_index.faiss

• keyword.py (Agent 1)

  • Implements the logic for analyzing and extracting key terms or topics from the user's query, which are then used to guide the document retrieval process.

• retrieve.py (Agent 2)

  • Contains the functionality for searching and retrieving relevant documents from the database. It implements the logic for querying and filtering the document set based on the user's input.

• rank.py (Agent 3)

  • Handles the ranking of the retrieved documents by relevance, ensuring the most pertinent documents are prioritized for response generation.

• generate.py (Agent 4)

  • Responsible for generating responses by synthesizing information from the top-ranked documents retrieved by the retrieve.py module. It composes concise answers based on relevant content.

Running the System

To run the system, execute the following command:

python -m pip install -r requirements.txt
python -m spacy download en_core_web_sm
streamlit run app.py