Deep Research Agent From Scratch

A comprehensive multi-agent research system built with LangGraph that can conduct intelligent research on any topic and generate detailed reports.

What It Does

This system transforms vague user queries into comprehensive research reports through a three-phase process:

• Scope → Clarify research requirements and generate structured briefs
• Research → Multi-agent parallel research with external tools
• Write → Synthesize findings into professional reports

Technical Skills Demonstrated

Agent Architecture & Patterns

• Multi-Agent Coordination: Supervisor pattern with parallel worker agents
• ReAct Agent Loops: Iterative research with tool calling and reflection
• State Management: Complex state flows across subgraphs and nodes

Advanced Integrations

• External APIs: Tavily search integration with content summarization
• LangGraph Workflows: End-to-end orchestration with conditional routing

Production-Ready Features

• Structured Output: Pydantic schemas for reliable AI decision making
• Async Orchestration: Strategic parallel processing for concurrent research
• Tool Ecosystem: Custom tools, MCP servers, and search optimization
• Error Handling: Robust workflow design with proper state transitions

Implementation Highlights

1. Smart Scoping - Clarifies ambiguous queries and generates detailed research briefs
2. Parallel Research - Multiple agents research different aspects simultaneously
3. Multi-Agent Supervision - Coordinator agent managing specialized research workers
4. Report Synthesis - Intelligent aggregation of research findings into cohesive reports

Key Technologies

• LangGraph for agent workflows and state management
• LangChain for LLM integration and tool orchestration
• Pydantic for structured output and data validation
• Async/Await for concurrent processing

Built a complete production-ready research system that showcases advanced agent patterns, multi-modal integrations, and sophisticated workflow orchestration.

Required API Keys

• OPENAI_API_KEY=your_openai_api_key_here
• TAVILY_API_KEY=your_tavily_api_key_here
• LANGSMITH_API_KEY=your_langsmith_api_key_here

How to Run

First install all the requirements from requirments.txt

Then, simply open and execute the full_agent.ipynb notebook:

jupyter notebook full_agent.ipynb