Retail Agent Learner

Pipeline Overview

This is an 8-agent pipeline for retail site selection analysis:

User Query
    ↓
IntakeAgent ────────────→ Extract location + business type
    ↓
MarketResearchAgent ────→ Google Search for demographics
    ↓
CompetitorMappingAgent ─→ Google Maps Places API
    ↓
GapAnalysisAgent ───────→ Python code execution (pandas)
    ↓
StrategyAdvisorAgent ───→ Extended thinking synthesis
    ↓
ParallelAgent ──────────→ Concurrent artifact generation
    ├── ReportGenerator ─→ HTML report
    ├── InfographicAgent → Image generation
    └── AudioOverview ───→ TTS podcast audio

What Each Agent Does

Key ADK Concepts

Agents

• LlmAgent: Single LLM call with tools and instructions
• SequentialAgent: Runs sub-agents one after another
• ParallelAgent: Runs sub-agents concurrently

State Flow

Data passes between agents via session state:

IntakeAgent → state["target_location"], state["business_type"]
MarketResearchAgent → state["market_research_findings"]
CompetitorMappingAgent → state["competitor_analysis"]
GapAnalysisAgent → state["gap_analysis"]
StrategyAdvisorAgent → state["strategic_report"]

Tools

Functions that agents can call to perform actions. Access state via ToolContext.

Callbacks

Lifecycle hooks: before_agent_callback and after_agent_callback.

Learning Path

Start with the 9-part tutorial series:

1. Part 1: Setup + First Agent
2. Part 2: IntakeAgent
3. Part 3: MarketResearchAgent
4. Part 4: CompetitorMappingAgent
5. Part 5: GapAnalysisAgent (code execution)
6. Part 6: StrategyAdvisorAgent (extended thinking)
7. Part 7: ArtifactGeneration (parallel outputs)
8. Part 8: Testing
9. Part 9: Production Deployment

[See references/architecture.md for detailed data flow] [See references/adk-concepts.md for ADK deep dive]