Agentic RAG Research Assistant

This repository demonstrates an Agentic Retrieval-Augmented Generation (RAG) approach to building a research assistant. It integrates a language model (invoked via Ollama), a vector database (Chroma), and specialized tools for retrieving, summarizing, comparing, and analyzing research papers.

---

Features

1. Tool-Based Retrieval

   • Uses a vector store (Chroma) to search large sets of research papers (PDFs or text).
   • Dynamically calls custom tools (summarize, compare, analyze citations).

2. Agentic Workflow

   • The language model decides if and when to invoke tools based on user queries.
   • Reduces hallucinations by fetching relevant context from external sources.

3. Conversational & One-Shot Modes

   • Conversational CLI: Engage with the assistant in a multi-turn conversation in your terminal.
   • One-Shot Method (agent.run(...)): Suitable for integration into web apps or other UIs (like Streamlit).

4. Streamlit UI

   • Basic web interface for entering a query and receiving a final answer.

---

File Overview

• requirements.txt
  Lists all Python dependencies (LangChain, Chroma, PyPDF2, etc.).

• config.py

  • OLLAMA_MODEL for specifying which Ollama model to use (default: "llama3.1:8b").
  • RESEARCH_PAPERS_DIR for the folder containing research papers (PDFs or text).

• ollama_client.py

  • A chat function that calls the Ollama CLI via subprocess.run, parsing JSON/text output.

• retriever.py

  • Loads and splits PDF/text research papers into chunks.
  • Builds a Chroma vector store to enable semantic similarity search.
  • Provides a retrieve_documents function to fetch relevant text from the corpus.

• tools.py

  • Defines tool functions that the language model can call:
    • get_search_results
    • summarize_paper
    • compare_papers
    • analyze_citations

• agent.py

  • Implements the agent logic:
    • Maintains a conversation (list of messages).
    • Decides when to invoke tools (based on the model’s JSON instructions).
    • Offers:
      • run(user_query) for one-shot queries.
      • converse() for a multi-turn, interactive conversation in the terminal.

• main.py

  • Entry point for the Conversational CLI. Initializes Agent and calls agent.converse().

• app.py

  • A Streamlit UI that provides a text input and displays the final answer returned by agent.run().

---

Setup

1. Install Dependencies

   pip install -r requirements.txt

2. Set Up the Ollama CLI

   • Install Ollama and ensure you have the desired model (e.g., llama3.1:8b) downloaded.
   • Verify it works by running:

     ollama run llama3.1:8b

   • Update OLLAMA_MODEL in config.py if you use a different model name.

3. Place Research Papers

   • Create a folder named research_papers in your project directory (or rename according to config.py).
   • Add .pdf or .txt files containing research papers you want to query.

4. (Optional) Provide OPENAI_API_KEY

   • If using OpenAI embeddings, you need an OPENAI_API_KEY.
   • Set it in your environment or pass it in your code:

     export OPENAI_API_KEY="your-openai-api-key"

   • Or switch to other embeddings from langchain_community if you prefer a local model.

---

Usage

1. Conversational CLI

Run the assistant in interactive mode:

python main.py

• Type your questions at the prompt.
• Type exit or quit to end.

2. One-Shot Query (as used by Streamlit)

If you just want a single answer, you can call:

final_answer = agent.run(user_query)

This is integrated in app.py, which uses Streamlit.

3. Streamlit UI

For a simple web-based UI:

streamlit run app.py

• A browser tab opens.
• Type a query and hit Submit to see a final answer.

---

FAQ

Q: I see Deprecation Warnings from LangChain.
A: Either update your imports (to langchain_community.*) or suppress warnings with:

import warnings
warnings.filterwarnings("ignore", category=DeprecationWarning)

We've already done this in some files.

Q: Why do I need OPENAI_API_KEY?
A: The default embeddings (OpenAIEmbeddings) call the OpenAI API to get vectors. If you prefer local embeddings, switch to another class from langchain_community.

Q: My model is calling unknown tool names.
A: We mapped synonyms in agent.py. If the model uses new/unexpected tool names, add them to self.tool_mapping.

Q: How do I add more features?
A: Just define new functions in tools.py and update agent.py. The LLM can call them if you prompt it with the correct tool schemas.

---

Contributing

Contributions and pull requests are welcome! For significant changes, please open an issue first to discuss the proposal. Feel free to fork this repo and customize it for your own research flows or embedding approaches.

---

Enjoy building and exploring your Agentic RAG Research Assistant!