Multi-Agent Research System
An AI research workflow that decomposes research into specialized agent tasks, gathers information from the web, extracts deeper source content, generates a structured report, and evaluates the final result through a dedicated critic agent.
Research Pipeline
Search → Read → Write → Critique
- Search Agent: Searches the web for recent and relevant information using dedicated research tools
- Reader Agent: Selects relevant sources and extracts deeper content from URLs
- Writer: Combines gathered research into a structured research report
- Critic: Reviews the generated report, scores its quality, identifies strengths, and highlights areas for improvement
Multi-Agent Architecture
Search Agent:
- Uses LangChain agent tooling for web research
- Performs web search and URL scraping
- Focuses on discovering relevant and recent sources
Reader Agent:
- Receives search results from the research stage
- Selects relevant sources for deeper inspection
- Uses URL scraping to extract additional context
Writer Chain:
- Combines search results and scraped content
- Generates structured research reports
- Produces findings, conclusions, and source references
Critic Chain:
- Evaluates the generated report
- Provides a quality score and structured feedback
- Identifies strengths and areas requiring improvement
Technical Architecture
- Agent Framework: LangChain
- LLM: Google Gemini
- Research Tools: Web Search, URL Scraping
- Pipeline: Search → Reader → Writer → Critic
- Application: Streamlit
- Language: Python
- Configuration: Environment-based API configuration
Key Capabilities
- Multi-Agent Research: Specialized agents handle different stages of the research process
- Web-Grounded Research: Research is based on dynamically discovered online sources
- Source Extraction: Relevant URLs are scraped for deeper information
- Structured Generation: Research is transformed into organized reports
- Self-Evaluation: Generated reports are independently reviewed by a critic stage
- Modular Pipeline: Research stages are separated into reusable agent and processing components
Stack
- Language: Python
- AI Framework: LangChain
- LLM: Google Gemini
- Agents: Search Agent, Reader Agent
- Tools: Web Search, URL Scraping
- Interface: Streamlit
- Output: Structured Research Reports