🔬 ResearchFlow.AI

An Autonomous Multi-Agent AI Research System

Type a topic. Four AI agents handle the rest.
Search the web → Scrape top sources → Write a structured report → Critique it for quality.

---

✨ What It Does

ResearchFlow is an agentic AI pipeline that automates deep research on any topic. Instead of one LLM call, it orchestrates four specialized agents in sequence — each with a distinct role — producing a polished, self-critiqued research report.

User Input ──▶ Search Agent ──▶ Reader Agent ──▶ Writer Chain ──▶ Critic Chain ──▶ Final Report
                  │                  │                 │                │
              Tavily API         Scrapes top        Structures      Reviews &
              web search           URL             findings       scores output

---

🤖 Agent Architecture

Agent	Role	Tool Used
Search Agent	Finds recent, reliable sources on the topic	Tavily Search API
Reader Agent	Picks the best URL and deep-scrapes its content	Web scraper tool
Writer Chain	Synthesizes all findings into a structured report	LLM chain (GPT / Claude)
Critic Chain	Reviews the report for accuracy, depth, and clarity	LLM chain (GPT / Claude)

Each agent is independently built and composed in pipeline.py, making the system easy to extend or swap individual components.

---

🗂️ Project Structure

researchflow-ai/
│
├── agents.py           # Agent + chain definitions (Search, Reader, Writer, Critic)
├── pipeline.py         # Core orchestration logic — wires agents into a sequence
├── tools.py            # Custom tool wrappers (search, scrape)
├── app.py              # Streamlit UI — real-time pipeline tracker + results viewer
│
├── assets/
│   └── screenshot.png  # UI screenshot for README
│
├── .env.example        # Template for required API keys
├── requirements.txt    # All dependencies
└── README.md

---

🚀 Quick Start

1. Clone the repo

git clone https://github.com/YOUR_USERNAME/researchflow-ai.git
cd researchflow-ai

2. Create a virtual environment

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

3. Install dependencies

pip install -r requirements.txt

4. Set up environment variables

cp .env.example .env

Open .env and fill in your API keys:

OPENAI_API_KEY=sk-...
TAVILY_API_KEY=tvly-...

5. Run the app

# Streamlit UI (recommended)
streamlit run app.py

# Or terminal mode
python pipeline.py

---

🖥️ UI Overview

The Streamlit interface is split into two panels:

Left panel — Control & Pipeline Tracker

• Topic input field
• Live step-by-step agent status (WAITING → RUNNING… → DONE ✓)

Right panel — Results

• Stat strip (word count, data ingested)
• Four tabs: Final Report · Search Results · Scraped Content · Critic Feedback
• One-click report download

---

🛠️ Tech Stack

Layer	Technology
Agent Framework	LangChain
Web Search	Tavily API
LLM Backend	OpenAI GPT-4o (swappable)
UI	Streamlit
Language	Python 3.10+

---

🔧 Configuration

You can swap the LLM backend in agents.py:

# Default — OpenAI
from langchain_openai import ChatOpenAI
llm = ChatOpenAI(model="gpt-4o")

# Switch to Anthropic Claude
from langchain_anthropic import ChatAnthropic
llm = ChatAnthropic(model="claude-3-5-sonnet-20241022")

---

📋 Requirements

langchain
langchain-openai
langchain-community
tavily-python
streamlit
python-dotenv

See requirements.txt for pinned versions.

---

🗺️ Roadmap

• Add memory / conversation history between runs
• Export reports as PDF
• Support multiple URLs per research run
• Add a confidence score to the critic output
• Deploy to Streamlit Cloud with one-click template

---

📄 License

MIT — free to use, modify, and distribute.

---

_{Built with LangChain · Streamlit · Python}