Multi-Agent Research System

A multi-agent pipeline that uses Claude to research any topic — planning queries, searching the web, analysing sources, synthesising findings, and generating a polished markdown report.

Architecture

Plan → Search (parallel) → Analyse (parallel) → Synthesise → Report

Agent	Model	Role
Planner	claude-opus-4-6	Decomposes topic into 3-4 targeted search queries
Search	claude-sonnet-4-6	Runs web searches and collects sources (parallel per query)
Analyst	claude-sonnet-4-6	Extracts structured findings from each source (parallel per source)
Synthesis	claude-opus-4-6	Cross-source reasoning, conflict resolution, gap identification
Report	claude-sonnet-4-6	Formats synthesis into a cited markdown report

Project Structure

├── main.py              # Entry point — orchestrates the pipeline
├── schemas.py           # Shared dataclasses (SearchResult, Finding, Synthesis, Report)
├── requirements.txt
└── agents/
    ├── __init__.py
    ├── search_agent.py
    ├── analyst_agent.py
    ├── synthesis_agent.py
    └── report_agent.py

Setup

# 1. Create and activate a virtual environment
python3 -m venv .venv
source .venv/bin/activate

# 2. Install dependencies
pip install -q "anthropic>=0.40.0"

# 3. Set your Anthropic API key
export ANTHROPIC_API_KEY="your-key-here"

Usage

# Run with a custom topic
python3 main.py "impact of large language models on software engineering"

# Default topic (if no argument provided)
python3 main.py

Output is saved to output/ as a markdown report and a JSON metadata file.