SmartStore AI — Master Build Roadmap (Phases 0–13)
This is the phase structure the full implementation guide follows. Each phase builds directly on the last — none of them are optional or reorderable, because each one depends on infrastructure the previous phase set up.
| Phase | Name | What gets built |
|---|---|---|
| 0 | Project Setup & Environment | Repo structure, Docker Compose (Postgres/Redis/Qdrant), FastAPI skeleton, SwiftUI skeleton, config/env handling |
| 1 | Database Design & Core Models | PostgreSQL schema, SQLAlchemy models, Alembic migrations for stores/products/users |
| 2 | Product Catalog Ingestion (RAG) | Chunking, embedding, Qdrant collection, the real ingest.py pipeline |
| 3 | Core RAG Query Pipeline | retrieve.py/query.py, grounded prompting, the first real /ask endpoint |
| 4 | SwiftUI MVP Chat UI | A real working chat screen calling /ask, rendering answers |
| 5 | Authentication | Firebase Auth on iOS + backend token verification |
| 6 | RBAC & Multi-Store Isolation | Roles, Postgres RLS, Qdrant metadata filtering by store |
| 7 | Agent Layer & Tool Calling | ReAct loop, real tools, multi-step reasoning |
| 8 | Streaming, Voice & Image Input | SSE streaming, on-device STT/TTS, photo-based product lookup |
| 9 | Caching & Cost Optimization | Redis session state, semantic caching, model routing |
| 10 | Observability & Evaluation | OpenTelemetry traces, Grafana, golden eval set, CI-gated evals |
| 11 | CI/CD & MVP Deployment | GitHub Actions, Dockerization, deploy to Render |
| 12 | Production Hardening & AWS | Secrets Manager, RDS, ElastiCache, S3, ECS Fargate, CloudWatch |
| 13 | Launch, Governance & Iteration | Governance docs, PostHog review loop, prompt A/B testing, continuous eval |
Why this order: Phases 0-4 get a thin, ungated, single-store vertical slice actually working end to end before anything else is added — that's deliberate, the same principle from the bootcamp's Volume 5 capstone. Auth and RBAC (5-6) come next because nothing past that point should be built without them. The agent layer and advanced input modalities (7-8) come before optimization (9) because you optimize a system that exists, not one you're still designing. Observability (10) comes before deployment (11-12) because you should never deploy something you can't see inside of. Governance (13) closes the loop once it's live.
Each phase will be delivered as its own complete document with full, real, runnable code — not snippets — plus a companion update to the actual project repository.
Status: All 14 phases (0–13) complete. ✅ 44 tests passing across the full suite. See docs/phase-13-launch-governance-guide.md for the closing summary and the full list of real bugs this build caught along the way.