A professional multi-agent AI system built with Google ADK, featuring LoopAgent for iterative research → draft → critique workflows, automated refinement, and content generation.
Built during the Google Cloud Technical Series: Multi-Agent AI Workshop
This repository demonstrates a multi-agent orchestration system using Google ADK.
The core component is the LoopAgent, which enables iterative workflows where agents:
- Research – collect relevant information (e.g., Wikipedia summaries).
- Draft – produce an initial content outline based on research.
- Critique – evaluate drafts, provide feedback, and decide whether to continue or stop the loop.
Once the iterative loop finishes, the system automatically writes the final draft to disk.
This workflow models real-world content generation pipelines, autonomous agent orchestration, and iterative refinement systems.
- Multi-Agent Workflow: Researcher, Screenwriter, Critic, File Writer agents.
- LoopAgent: Iterative research → draft → critique cycle.
- SequentialAgent: Executes ordered steps after iterative loop.
- Automated File Output: Drafts are saved in
movie_pitches/folder. - State Management: Accumulates research, drafts, and feedback in
session.state. - Extensible: Replace mock screenwriter with LLM-based models easily.
- Portfolio Ready: Demonstrates agentic AI workflow, suitable for recruiter showcase.
By building this system during the Google Cloud Technical Series, I gained hands-on experience in:
- Designing multi-agent pipelines for content generation
- Implementing LoopAgent and SequentialAgent workflows
- Managing agent state, persistence, and feedback loops
- Integrating deterministic tools (APIs, file writing) with creative LLM outputs