How vessl-ai/ai-agent-research works
Overall Structure
• That this is a heterogeneous collection of smaller projects (i.e. modules) focusing on different agent-based research tasks.
Main Components
• Individual Projects (modules):
• CrewAI Process Type Comparison (crewai-process-type)
• Agentic RAG System (agentic-rag)
• Code Agent (code-agent)
• Agentic Crawl (agentic-crawl)
• CrewAI Deep Research (crewai-deepresearch)
• Configuration Component: Shows YAML files (e.g., agents.yaml, tasks.yaml) and environment templates (.env.template) for each project.
• Core Application Logic: Represented by main source files (e.g., main.py, run.py, app.py) that drive the business logic.
• Auxiliary Tools or Agents: For instance, code-agent leverages tools for code analysis; agentic-crawl has a crawl_agent; deep research project has its own deep_research_agent.
• Presentation/Output Layer: Some projects have a “static” directory (with index.html) or produce HTML reports, indicating output rendering.
• Data/Knowledge Stores: Project files (like PDFs in agentic-rag’s knowledge folder and research result documents) show where inputs or outputs are stored.
Architectural Patterns and Design Principles
• Modular nature – each project is self-contained but contributes to the greater research efforts.
• Separation of concerns:
• Configuration is externalized via YAML and .env templates.
• Business logic resides in Python source files.
• Presentation/output is clearly separated (static files or generated reports).
• If applicable, note any patterns like orchestration of agents or microservice-like structures (though they are not full-blown microservices, the projects operate independently and may even interact).
Specific Inclusions
• Each module should have a visible “Configuration Layer” where environment variables and YAML settings are read.
• The “Application Layer” in each should clearly show which file (like main.py or run.py) triggers the logic.
• For agent execution, show components like:
• Retrieval-Augmented Generation in agentic-rag: illustrate two paths for PDFSearchTool and CrewDoclingSource, complete with document ingestion.
• Code Analysis Agent in code-agent: include an element that processes source code and outputs HTML reports.
• Process Comparison in crewai-process-type: denote performance comparison between sequential and hierarchical orchestrations.
• Data flow from the input (e.g., PDFs, source code) through processing agents.
• Outputs being generated (reports in HTML or result text files).