[FEAT] Story 2.1 Dynamic Scenario-to-Prompt Engine

[yeomin4242]

한줄 설명

Create the FastAPI service that converts scenario JSONB parameters into Gemini 2.5 Flash system prompts, maintaining character consistency across multiple scenarios.

문제·기회

No response

제안 내용

No response

완료 기준(AC)

• /api/ai/adapt-prompt endpoint accepts scenario_id and base_prompt
• Scenario type-specific templates: CHARACTER_CHANGE, EVENT_ALTERATION, SETTING_MODIFICATION
• JSONB parameter interpolation into prompt templates
• Gemini 2.5 Flash system_instruction format (optimized for Gemini API):
  • Uses Gemini's system_instruction parameter (not prepended to user message)
  • Token-efficient prompt structure (reduces input token cost)
  • Supports Gemini's contents array for multi-turn context
• Character consistency layer preserves core traits across scenarios (e.g., Hermione's intelligence maintained)
• Negation instruction for altered properties (e.g., "Hermione is NOT in Gryffindor")
• Meta-scenario handling for forked scenarios (combines parent + child parameters)
• VectorDB character retrieval: Query ChromaDB for character personality traits (768-dim embeddings)
• Prompt caching with Redis (TTL 1 hour) for identical scenario_id + base_prompt
• Gemini API retry logic: 3 attempts with exponential backoff (1s, 2s, 4s)
• Response time < 50ms (cached), < 500ms (uncached with VectorDB lookup)
• Unit tests >80% coverage

관련 참고자료

No response

관련 이슈·블로커

No response