Synth — AI Research Agent Platform

A multi-agent AI research platform that generates deep research reports with autonomous agents for search, content extraction, writing, and critique.

Overview

Synth orchestrates four specialized AI agents working in sequence:

1. Search Agent - Queries Tavily for recent, reliable sources
2. Reader Agent - Scrapes and extracts content from URLs using BeautifulSoup
3. Writer Chain - Synthesizes research into structured reports
4. Critic Chain - Reviews and scores reports with improvement suggestions

Project Structure

├── backend/
│   ├── main.py          # FastAPI entry point
│   ├── agents.py        # Agent definitions (search, reader, writer, critic)
│   ├── tools.py         # Tools for web search and URL scraping
│   ├── pipeline.py      # Main orchestration pipeline
│   ├── requirements.txt # Python dependencies
│   └── .env             # Backend API keys (TAVILY_API_KEY, GROQ_API_KEY, API_BASE, ALLOWED_ORIGINS)
│
└── frontend/
    ├── src/
    │   ├── app/
    │   │   └── page.tsx          # Root page — renders LandingPage or Workspace
    │   ├── components/
    │   │   ├── LandingPage.tsx   # Landing / hero screen
    │   │   ├── Workspace.tsx     # Main research workspace with agent graph
    │   │   └── AgentNode.tsx     # Individual agent status card
    │   └── lib/
    │       ├── api.ts            # API client (calls FastAPI backend)
    │       └── markdown.ts       # Markdown → HTML + score extraction helpers
    ├── package.json
    ├── tailwind.config.js
    ├── next.config.js
    └── .env                      # NEXT_PUBLIC_API_URL=http://127.0.0.1:8000

Tech Stack

Backend

• FastAPI - Backend web framework
• LangChain - Agent framework and chains
• Groq API - LLM inference
• Tavily API - Web search engine
• BeautifulSoup - HTML content extraction
• Python 3.x - Backend runtime

Frontend

• Next.js 14 - React framework (App Router)
• TypeScript - Type-safe components
• Tailwind CSS - Utility-first styling
• React Hooks - State and side-effect management

Setup

Backend

# From the project root
cd backend

# Create and activate virtual environment
uv venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate

# Install dependencies
uv pip install -r requirements.txt

Frontend

cd frontend

# Install Node dependencies
npm install

Configuration

backend/.env

TAVILY_API_KEY=your_tavily_key
GROQ_API_KEY=your_groq_key
API_BASE=http://127.0.0.1:8000
ALLOWED_ORIGINS=http://localhost:3000

frontend/.env

NEXT_PUBLIC_API_URL=http://127.0.0.1:8000

Important: NEXT_PUBLIC_API_URL must point to the FastAPI backend port (8000), not the Next.js dev server port (3000).

Running the Application

Terminal 1 — Start the backend

cd backend
source .venv/bin/activate
uvicorn main:app --reload --port 8000

Terminal 2 — Start the frontend

cd frontend
npm run dev

Open http://localhost:3000 in your browser.

API Endpoints

• POST /research - Run the full research pipeline for a given query
• GET /config - Returns runtime config (API base URL)
• GET / - Health check

Architecture

Next.js Frontend ←→ FastAPI (main.py) ←→ run_research_pipeline (pipeline.py)
                                                      ↓
                        ┌────────────────────────────────────────────────────┐
                        │  Agent Execution Layer (agents.py)                 │
                        │  ├─► build_search_agent() → web_search (tools.py) │
                        │  ├─► build_reader_agent() → scrape_url (tools.py) │
                        │  ├─► writer_chain (LLM)                           │
                        │  └─► critic_chain (LLM)                           │
                        └────────────────────────────────────────────────────┘

Clean Separation:

• API Layer (main.py) - HTTP endpoints and CORS middleware
• Research Orchestration (pipeline.py) - Research workflow coordination
• Agent Execution (agents.py) - Agent builders and LLM chains
• Tools (tools.py) - Web search and scraping implementations
• Frontend (Workspace.tsx) - Multi-agent UI with live activity log and progress graph

Data Flow:

1. User enters a query in the Next.js frontend
2. lib/api.ts POSTs the query to POST /research on the FastAPI backend
3. FastAPI calls run_research_pipeline() synchronously and returns the full result
4. Frontend animates agent states (Search → Reader → Writer → Critic) as the result arrives
5. Report, critic feedback, and sources are rendered in the Workspace UI

Features

• Deep web research with source verification
• Multi-agent workflows with live agent graph
• Inline citations and source tracking
• Quality scoring (out of 10) with critic feedback
• Report export (plain text download) and clipboard copy
• Quick, Deep, and Report-only research modes
• Collapsible sidebar with research history

Research Examples

• AI agent frameworks comparison
• Electric vehicle market trends
• LLM prompt engineering best practices
• Startup fundraising landscape
• Quantum computing applications