RFE Builder Phase 2: Conversational AI Interface

[jeremyeder]

Phase 2: Conversational AI Interface Implementation

Overview

Transform RFE Builder from form-based to conversational AI interface using Anthropic Claude API with sophisticated prompt management system.

Core Features

1. Role-Specific Prompt Templates

• 7 agent-specific prompt templates (Parker, Archie, Stella, Olivia, Lee, Taylor, Derek)
• Workflow step-aware prompting
• Standardized template management system

2. Cost Optimization

• Token usage tracking and optimization
• Context pruning (relevant RFE sections only)
• Response caching for identical inputs
• Progressive disclosure (simple → complex prompts)
• Structured JSON outputs

3. Dynamic RFE Builder Integration

• Conversational RFE creation replaces forms
• Context-aware prompt selection based on RFE state
• Historical pattern matching for recommendations
• Missing information detection and guided collection

Technical Architecture

File Structure

/demos/rfe-builder/
├── components/
│   ├── chat_interface.py      # Claude-powered chat UI
│   ├── ai_assistants.py       # 7 agent-specific helpers
├── ai_models/
│   ├── prompt_manager.py      # Hybrid template management
│   ├── cost_tracker.py        # Token optimization
│   ├── rfe_classifier.py      # Dynamic prompt selection
├── prompts/
│   ├── agents/
│   │   ├── parker_pm/
│   │   │   ├── step1_prioritization.yaml
│   │   │   └── step6_communication.yaml
│   │   ├── archie_architect/
│   │   │   ├── step2_technical_review.yaml
│   │   │   └── step4_acceptance_criteria.yaml
│   │   └── [other agents...]

Prompt Management Strategy

• Enum-based mapping for type safety
• Workflow-aware templates per agent per step
• Context-sensitive selection based on RFE state
• Maintainable organization by agent ownership

Dependencies

• anthropic>=0.25.0 - Claude API integration
• streamlit-chat>=0.1.0 - Enhanced chat UI
• pyyaml>=6.0 - Template management
• tiktoken>=0.5.0 - Token counting

Success Criteria

• All 7 agents have functional AI assistance
• Conversational RFE creation replaces form interface
• Cost tracking with token optimization implemented
• Dynamic prompting works based on RFE context
• Maintains all Phase 1 functionality
• Comprehensive test coverage for AI features

Future Enhancements (Separate Issues)

• Enterprise prompt standardization framework
• AI performance monitoring dashboard
• Multi-tenant prompt management
• Advanced workflow intelligence with ML predictions

Labels

• enhancement
• phase-2
• conversational-ai
• anthropic-claude
• prompt-management

[jeremyeder]

fixed by #20