🧠 Agentic RAG Research System

An autonomous multi-agent research pipeline that plans, retrieves, analyzes, and synthesizes technical documentation — powered by cloud-hosted LLMs and real-time web retrieval.

Live Demo: https://deepresearcher.streamlit.app/

---

🎯 Project Objective

This system emulates a professional research workflow by combining planning, retrieval, and reasoning capabilities into a fully automated pipeline.

• 🔍 Decompose a research topic into structured, targeted questions
• 🌐 Retrieve real-time information from the web via Tavily Search API
• 🧠 Reason over retrieved context using cloud-hosted LLMs
• 📄 Generate high-quality, structured technical documentation
• 📤 Export research outputs in TXT, PDF, and Word formats

Built as part of the OpenDeepResearcher: Agentic LLM Research Framework internship project.

---

🧩 System Architecture

The application follows a sequential multi-agent workflow, where each agent operates as an independent module with a clearly defined responsibility.

┌─────────────────────────────────────────────────────────────┐
│                     USER INPUT (Topic)                      │
└─────────────────────┬───────────────────────────────────────┘
                      │
                      ▼
┌─────────────────────────────────────────────────────────────┐
│              🧩 PLANNER AGENT  (planner_agent.py)           │
│  Analyzes the topic and generates targeted research         │
│  questions. Acts as the strategic blueprint for the         │
│  entire research process.                                   │
└─────────────────────┬───────────────────────────────────────┘
                      │  Research Questions
                      ▼
┌─────────────────────────────────────────────────────────────┐
│              🔍 SEARCHER AGENT (searcher_agent.py)          │
│  Uses Tavily Search API for real-time web retrieval.        │
│  Collects contextual facts and reference sources to         │
│  supply grounded information to the Writer Agent.           │
└─────────────────────┬───────────────────────────────────────┘
                      │  Context + Sources
                      ▼
┌─────────────────────────────────────────────────────────────┐
│              ✍️  WRITER AGENT   (main.py)                    │
│  Synthesizes planner questions and retrieved context        │
│  into a professional research report with structured        │
│  sections: Executive Summary, Technical Analysis,           │
│  Challenges & Ethics, and Strategic Conclusion.             │
└─────────────────────┬───────────────────────────────────────┘
                      │  Final Report
                      ▼
┌─────────────────────────────────────────────────────────────┐
│              📤 EXPORT (TXT / PDF / Word)                   │
└─────────────────────────────────────────────────────────────┘

---

🛠️ Tech Stack

Component	Technology	Purpose
LLM Inference	Groq (Llama 3.1 8B)	Cloud-hosted, OpenAI-compatible LLM API
Web Search	Tavily Search API	Real-time information retrieval
Frontend	Streamlit	Interactive web UI with dark/light mode
PDF Export	ReportLab	Professional PDF document generation
Word Export	python-docx	Word document generation

---

📁 Project Structure

├── app.py                 # Streamlit UI — interactive research workflow
├── main.py                # Writer Agent + CLI entry point
├── planner_agent.py       # Planner Agent — generates research questions
├── searcher_agent.py      # Searcher Agent — retrieves web context via Tavily
├── llm_utils.py           # Shared utilities — LLM calls (Groq) & Tavily search
├── requirements.txt       # Python dependencies for deployment
├── .gitignore             # Protects .env and secrets from GitHub
├── .env                   # API keys (local only, never committed)
└── README.md              # Project documentation

---

⚡ Quick Start

Prerequisites

• Python 3.10+
• Groq API Key (free)
• Tavily API Key (free tier available)

1. Clone the Repository

git clone https://github.com/dinesh9997/OpenDeepResearcher-Agentic-LLM-Research-Framework.git
cd OpenDeepResearcher-Agentic-LLM-Research-Framework

2. Install Dependencies

pip install -r requirements.txt

3. Configure Environment Variables

Create a .env file in the project root:

LLM_API_KEY=your_groq_api_key_here
LLM_BASE_URL=https://api.groq.com/openai/v1
LLM_MODEL_NAME=llama-3.1-8b-instant
TAVILY_API_KEY=your_tavily_api_key_here

4. Run the Application

streamlit run app.py

The app will be available at http://localhost:8501

---

🎨 User Interface

• Dark Mode & Light (Gold) Mode toggle
• Clean, card-based layout with gradient animations
• Guided research flow: Topic → Planner → Writer → Export
• One-click export to TXT, PDF, or Word

---

🚀 Deployment (Streamlit Cloud)

1. Push this repository to GitHub
2. Go to share.streamlit.io
3. Connect your repository → Branch: main
4. Set Main file path: app.py
5. Add secrets in Settings → Secrets:

LLM_API_KEY = "gsk_your_groq_key"
LLM_BASE_URL = "https://api.groq.com/openai/v1"
LLM_MODEL_NAME = "llama-3.1-8b-instant"
TAVILY_API_KEY = "tvly_your_tavily_key"

6. Click Deploy 🚀

---

📤 Export Formats

Format	Library	Features
📄 TXT	Built-in	Plain text with markdown formatting
📄 PDF	ReportLab	Professional layout with styled headings
📄 Word	python-docx	Editable document with paragraph structure

---

🔒 Security

• API keys are stored in .env (local) or Streamlit Secrets (cloud)
• .gitignore ensures .env is never pushed to GitHub
• No hardcoded credentials anywhere in the codebase

---

✅ Internship Alignment

This project fully aligns with the internship objectives by:

• ✅ Implementing a complete agentic research pipeline
• ✅ Using cloud-hosted LLMs via Groq for scalable deployment
• ✅ Integrating real-time web retrieval via Tavily Search API
• ✅ Delivering a polished UI with professional export capabilities
• ✅ Following modular architecture with separated agent responsibilities
• ✅ Demonstrating end-to-end system design and execution

---

👤 Author

GUJJU DINESH B.Tech – CSE (Artificial Intelligence)

---