Core Engine Build Status
Epic-level progress tracking for the GamiWays Core Engine build — manually synced from the development repository.
gami-lab/gami-digidouble-corePhase A — Minimal Core
Build and validate the fundamental loop: user input → context assembly → orchestrated avatar response → memory update. Deliver a usable back-office and text-based prototype.
Monorepo (pnpm + Turborepo), TypeScript strict, Docker Compose, CI baseline
Direct provider SDKs, streaming abstraction, JSON enforcement, retry/timeout wrappers
Session + Conversation entities, message routing, exchange persistence
PostgreSQL + pgvector schema, Redis session cache, ioredis client
Back-office Next.js, session inspector, conversation replay, GM debug view
Scenario config CRUD, avatar management, API key auth
Game Master state inspector, directive log, async trigger trace
Session-level state visualization, memory snapshot, context assembly trace
Health endpoints, dependency checks (DB, Redis, LLM), structured logging
Director–Actor model, async trigger evaluation, GM directive injection into avatar flow
Working memory (sliding window + cumulative summary), episodic persistence, long-term user facts, deterministic selection policy, memory hydration
Latency measurement (TTFT, end-to-end), token accounting, cost tracking per session/scenario
Avatar switching logic, GM-driven transitions, persona isolation per conversation
ObservedLlmAdapter enforced across all call sites — Langfuse traces, latency, token counts, cost estimates, model metadata
Document ingestion (PDF, MD, text), chunking, embeddings, pgvector retrieval, context-aware filtering
3-dimension context assembly: Memory + Experience/World + Knowledge. Deterministic injection policy.
WebSocket streaming, SSE fallback, progressive token delivery to frontend
Scenario objectives tracking, progression state machine, GM-driven narrative milestones
Scenario configuration UI, avatar assignment, knowledge source binding, objective definition
End-to-end test with a real use case (learning or storytelling), user feedback collection
Functional prototype: text-based, back-office operational, one real scenario deployed
Phase B — Enhanced Experiences
Add voice input/output, multimedia triggers, multiple scenarios, richer memory systems, and a user-facing frontend.
STT pipeline (Deepgram / Whisper), TTS streaming (Cartesia / Inworld TTS-2), real-time audio delivery
GM-driven media events: image display, video playback, document reveal during conversation
Conversation UI, session history, progression visualization, multi-device support
Phase C — Research & Scale Readiness
Prepare the platform for advanced integrations: expressive avatars, advanced persona systems, scaling, SDKs, and research partnerships.
Real-time video avatar binding, body language triggers, lip-sync coordination
Tenant isolation, JWT auth, RBAC, data residency controls
Public SDK, versioned API, developer documentation, integration examples