DeepSearch-Agent

A Multi-step autonomous research agent. It takes a high-level user query, interactively refines the research scope, dynamically plans a report outline, gathers information from the web, and writes a comprehensive, cited report in Markdown

It uses Tavily API for web search/scraping and Gemini AI API for LLM.

Live Demo

Deployed on Hugging Spaces

The UI is barebones, but it works.

Features

• Human-in-the-Loop: Starts by asking clarifying questions to narrow down the user's intent.
• Dynamic Outline Planning: Generates a structured report outline based on initial search results, then expands each section with key questions.
• Deep Research: Performs targeted, deep-dive searches for each section of the report.
• Retrieval-Augmented Generation (RAG): Chunks and embeds research content into a vector store (FAISS) to find the most relevant information for writing.
• Source Citation: Meticulously cites every factual statement, linking it back to the source URL.
• Context-Aware Writing: Keeps track of previously written sections to maintain flow and avoid repetition.
• PDF Export: Converts the final Markdown report into a high-quality, well-formatted PDF with a table of contents using Pandoc and LaTeX.

Getting Started

Prerequisites

• Python 3.8+
• Pandoc: Required for converting the markdown report to PDF.
• A LaTeX distribution, such as MiKTeX (for Windows) or TeX Live (for macOS and Linux). This is required by Pandoc to create PDFs.

Installation

1. Clone the repository:

   git clone https://github.com/rajput-musa/DeepResearch-Agent.git
   cd DeepResearch-Agent

2. Create a virtual environment and activate it:

   python -m venv venv
   venv\Scripts\activate  # On Windows
   # source venv/bin/activate  # On macOS/Linux

3. Install the required Python packages:

   pip install -r requirements.txt

4. Set up your environment variables:

   Create a file named .env in the root of the project directory and add your API keys:

   GOOGLE_API_KEY="YOUR_GOOGLE_API_KEY"
   TAVILY_API_KEY="YOUR_TAVILY_API_KEY"
   

   You can get your API keys from:

   • Google AI Studio
   • Tavily

Running the Application

Once you have completed the installation and setup steps, you can run the application with the following command:

python app.py

This will start the Gradio web server. Open the provided URL in your browser to start using the Mini-DeepSearch-Agent.

How it Works

1. Initial Topic: You provide a research topic.
2. Clarification: The agent asks clarifying questions to narrow down the scope and understand your requirements.
3. Research & Report Generation: Based on your answers, the agent conducts research and generates a report section by section. You can see the progress in the UI.
4. Download Report: Once the report is complete, a "Download Report as PDF" button will appear. Click it to download the report.

Project Structure

.
├── .env.example
├── app.py
├── research_agent
│   ├── agent.py
│   ├── config.py
│   ├── export.py
│   ├── prompts.py
│   ├── rag_pipeline.py
│   └── tools.py
├── LICENSE
└── requirements.txt