HotelMind AI
Product
Product Roadmap
Phase-by-phase delivery status and a Now/Next/Later plan.
Source: docs/product-management/11-product-roadmap.md
HotelMind AI — Product Roadmap
Scope: A phased roadmap reconciled against actual delivery status, not aspirational planning. Phases use delivery state rather than committed calendar dates, except where an indicative long-horizon view is explicitly labeled as such.
1. Phases
| Phase | Objective | Status |
|---|---|---|
| Foundation | Core operational data model and platform architecture | Delivered |
| Hotel Operational System | Restaurant/staff modules, review ingestion, full dashboard UX | Delivered for real modules; partial (mock-backed) for operational-preview modules |
| Data Engineering | Airflow-orchestrated dbt star-schema warehouse on Postgres, MinIO data lake | Delivered |
| AI/ML | Forecasting and recommendation models across revenue, restaurant, staff, and guest domains | Delivered, with documented limitations (synthetic training data on two models, weak occupancy accuracy, churn label leakage) |
| Generative AI | Conversational AI assistant and rule-based operational insights | Delivered in code; degraded in production (LLM generation disabled, falls back to retrieval-only) |
| MLOps | Model registry, experiment tracking, drift monitoring, promotion/rollback | Delivered for internal engineering use; not yet business-facing |
| Real-Time Analytics | Kafka event streaming and WebSocket push to the dashboard | Delivered |
| Cloud/Infra Architecture | Docker Compose on a single VM, Nginx, GitHub Actions CI/CD | Delivered for current scale; horizontal scaling and disaster recovery are planned |
| Security/Compliance | SSO/MFA, audit logging, formal compliance posture | Planned, not started |
| Executive Dashboard | Cross-domain reporting with data-source transparency | Delivered for real modules; mock-to-real migration ongoing |
| Digital Twin | "What-if" operational simulation | Future / not started — depends on forecasting models being validated on real data first |
2. Now / Next / Later
Now — protect what's shipped
- Restore generative AI to production reliability.
- Replace the hardcoded room-capacity constant with live inventory.
- Re-baseline the churn model without label leakage.
These are corrections to already-shipped, customer-visible claims. Shipping further on top of a known-flawed baseline compounds risk.
Next — enterprise credibility + mock-to-real migration
- SSO/MFA, audit logging.
- Begin converting mock-backed modules (housekeeping, maintenance, check-in/out) to real backend integration.
These unlock enterprise trust and reduce the real/mock gap, but are not blocking current functionality.
Later — new capability, higher uncertainty
- Real POS/order-data integration and retraining.
- Shift-schedule optimizer.
- Portfolio (multi-property) real backend.
- LSTM occupancy model as an alternative approach.
- Digital Twin.
Each depends on external integration, additional real data, or has low current confidence relative to effort.
3. Delivered vs. Planned vs. Future Vision
| Category | Items |
|---|---|
| Delivered | Foundation, core operations, data warehouse, all 5 ML models (with documented limitations), RAG retrieval, AI Insights, real-time analytics, /reports, data-source badges, single-VM deployment |
| Planned | Live-inventory capacity fix, churn re-baseline, generative AI restoration, SSO/MFA, audit logging, mock-to-real migration |
| Future Vision | Real POS/order data integration, shift optimizer, real multi-property architecture, LSTM occupancy model, Digital Twin |