Research-Assistant

An AI-powered research assistant that performs multi-step, deep research on any topic by combining web search (via Tavily) and large language model-based agents.

The goal of this repository is to provide a clean and understandable implementation of a deep research agent system using LangChain and LangGraph. It showcases how multiple specialized agents can collaborate to perform credible research, draft structured answers, refine them, and fact-check information.

How It Works

The system is based on a multi-agent pipeline, where each agent specializes in a part of the research and answer development process:

• A Research Agent fetches data from the web using Tavily.
• A Drafting Agent organizes research into a structured response.
• Depending on the content, either a Refinement Agent improves unclear drafts, or a Fact-Checking Agent verifies controversial claims.
• Finally, a Polishing Agent performs a last refinement pass before delivering the final answer.

The entire agent workflow is orchestrated dynamically using LangGraph.

Features

• Online Research: Gathers fresh web data using Tavily search API.
• Answer Drafting: Summarizes and organizes research into clear, professional responses.
• Refinement and Fact-Checking: Automatically routes drafts for improvement or fact-checking based on detected needs.
• Final Polishing: Every answer is finalized through an editing agent for clarity and professionalism.
• Dynamic Routing: Uses LangGraph conditional routing to adapt the workflow based on research content.
• Streamlit Frontend: Clean web app interface with chat history and a minimalistic user experience.

Requirements

• Python 3.10+
• API keys for:
  • Tavily Search API
  • Groq LLM API (or another supported LLM)

Setup

Python Environment

1. Clone the repository:

   git clone https://github.com/your-username/deep-research-assistant.git
   cd deep-research-assistant

2. Install dependencies:

   pip install -r requirements.txt

3. Create a .env file and set your API keys:

   TAVILY_API_KEY="your_tavily_key"
   GROQ_API_KEY="your_groq_key"
   

4. Run the application:

   streamlit run app.py

Docker (Optional)

1. Clone the repository.

2. Create and configure .env with your keys.

3. Build and run the Docker container:

   docker build -t deep-research-assistant .
   docker run -p 8501:8501 deep-research-assistant

Usage

Once running, the app will:

• Let you enter a research query.
• Automatically:
  • Search for information.
  • Draft an answer.
  • Refine or fact-check the answer as needed.
  • Finalize the output.
• Display the final answer on the main page.
• Store previous queries and answers in the sidebar under "Chat History."

Agent Architecture

• ResearchAgent: Uses Tavily search to gather information.
• AnswerDrafter: Converts research into an initial draft.
• RefinerAgent: Enhances unclear or incomplete drafts.
• FactCheckerAgent: Verifies controversial information based on research.
• FinalPolishAgent: Improves overall clarity, structure, and professionalism.
• Controller (LangGraph): Decides whether to refine, fact-check, or finalize the draft based on heuristics.

System Workflow

1. Research agent fetches online information.

2. Drafting agent organizes findings into a structured answer.

3. Routing logic checks the content:

   • If unclear → Send to RefinerAgent.
   • If controversial → Send to FactCheckerAgent.
   • Otherwise → Proceed.

4. All paths finish with a final polishing step.

5. The final answer is shown in the UI and stored in chat history.

   

   Output with Streamlit

   

To-Do and Improvements

• Add multi-turn conversation or follow-up question capability.
• Support additional LLM backends beyond Groq.
• Implement more fine-grained control over research depth and breadth.