Implement Workflow and LLM Configuration

[Haydart]

Task 009: Implement Workflow and LLM Configuration

Description

Create TrendDetectionWorkflow that wires TokenLaunchSource → TrendDetectionProcessor → TrendFirebaseExecutor, and create trend_llm.yaml configuration.

Acceptance Criteria

• Create TrendDetectionWorkflow in src/workflows/
• Wire TrendDetectionSource (custom source with Firestore watch)
• Wire TrendDetectionProcessor
• Wire TrendFirebaseExecutor
• Implement event loop: TrendDetectionEvent → process → execute
• Create src/res/config/trend_llm.yaml configuration file
• Configure text_model_name: claude-3-haiku (classification)
• Configure multimodal_model_name: claude-3.5-sonnet (creation)
• Set text_temperature: 0.1, multimodal_temperature: 0.3
• Define fallback models for both text and multimodal
• Add environment variable defaults (TREND_CACHE_TTL_SECONDS=300, TREND_PROMOTION_THRESHOLD=5)

Technical Details

• Workflow location: src/workflows/trend_detection_workflow.py
• Config location: src/res/config/trend_llm.yaml
• Use TrendDetectionSource (wraps TokenLaunchSource + Firestore watch)
• Event flow: TrendDetectionEvent (enriched) → processor.process() → executor.execute(action)
• LLM config format matches existing litellm.yaml structure
• Text model: fast/cheap for classification (90% of cases)
• Multimodal model: expensive for creation (10% of cases)

CRITICAL: Workflow instantiates TrendDetectionSource with:

• websocket_source = TokenLaunchSource() (existing)
• firebase_service = SimpleFirebaseService() (existing)

This gives the workflow real-time trend updates with zero additional infrastructure.

Effort Estimate

• Size: M
• Hours: 6-8
• Parallel: false (depends on Tasks 003, 007, 008 - UPDATED)