AI Expert Recommendations Summary & Strategic Direction

[anchapin]

Summary of AI Expert Analysis from conversation-1752371882583.json

Current Repository Status Assessment

Direction score: 8/10 (Technical), 6/10 (Velocity/focus)

• ✅ Architecture choices align perfectly with PRD
• ✅ Clean Python FastAPI backend, React frontend, CrewAI agents
• ✅ Docker-compose dev env, GitHub Actions CI
• ⚠️ Need "vertical slice" - actual end-to-end conversion

Key Technical Gaps Identified

1. No real Java AST parsing yet (🔴 Blocked for MVP)
2. Hard-coded block template (🟡 Limits functionality)
3. Missing texture pipeline (🟡 Texture not copied)
4. No .mcaddon packager (🔴 Cannot test on Bedrock)
5. Frontend shows "Coming Soon" (🟡 Blocks user feedback)
6. No sample .jar in repo (🟡 Hard to test)

Strategic Recommendations

RL/AI Approach

• Skip RL for MVP - Current CrewAI + templates sufficient
• Start collecting data - Scrape CurseForge for training data
• Add lightweight reward simulator - Unit tests, schema validation
• Revisit RL in Phase 4 - After 1k+ labeled trajectories

7-Day Sprint Priority

Focus on vertical slice proving architecture works end-to-end

Future Considerations

• Complex mod conversion has hard ceiling (~20-60% success rate)
• Tool should be "smart assistant" not "perfect converter"
• Best value: accelerate manual porting workflow

Related Issues: #167, #168, #169, #170, #171, #172, #173, #174