Feature Matrix & Product Roadmap¶
Museum AR Smart Glass Guide · HumanityAI'd — July 2026
Confidential
Download
1. Feature matrix (v1.0 — shipping today)¶
| Capability | Status | Notes |
|---|---|---|
| Hands-free exhibit recognition (CLIP + FAISS) | ✅ Live | Verified 0.995 similarity on enrolled exhibits |
| Multi-angle enrolment (1–3 photos/exhibit) | ✅ Live | Live index update, no retraining |
| 8 languages (en, ar, fr, es, zh, ru, hi, de) | ✅ Live | Engine supports 23 |
| Arabic + RTL content | ✅ Live | name_ar fields, RTL data model |
| Neural TTS narration (Chatterbox multilingual) | ✅ Live | Museum-tuned loudness, Opus |
| Human review/approval of narration | ✅ Live | Arabic requires explicit approval |
| Dwell-based auto-play ("look & listen") | ✅ Live | Tunable trigger/cooldown |
| WebRTC live streaming from glasses | ✅ Live | H.264, DataChannel results |
| Snapshot recognition (fallback) | ✅ Live | POST /recognize |
| Guided tours (ordered stops) | ✅ Live | Multilingual tour metadata |
| Sessions & engagement analytics | ✅ Live | Dwell, play-through, popular exhibits |
| Admin content panel | ✅ Live | CRUD museums/galleries/exhibits/translations |
| On-prem Docker deployment | ✅ Live | Single GPU host, no cloud |
| Prometheus/Grafana ops | ✅ Live | Metrics + dashboards |
| Glasses settings (web panel) | ✅ Live | Display/config for the fleet |
2. Known limitations (honest)¶
- Single-node / single-worker backend — one GPU box ≈ one museum, capped concurrent streams.
- Narration generation is an admin/offline activity (GPU-intensive); runtime serving uses pre-generated audio.
- Authentication and transport security are POC-grade — hardening scheduled (see Due-Diligence report).
- WebRTC streaming path pending on-hardware capacity verification; snapshot path fully proven.
3. Roadmap¶
v1.1 — Hardening (0–3 months)¶
- JWT validation + role-based access; per-device provisioned tokens; TLS + pinning on glasses
- Background TTS job queue; thread-pooled inference (unblock event loop)
- Automated tests + CI on both repos; dependency CVE upgrades
- FAISS rebuild-from-DB durability;
museum_idmulti-tenant filtering
v1.5 — Reach & scale (3–6 months)¶
- BYOD mode: visitor's own phone as an alternative to glasses (widen adoption)
- Horizontal scale: stateless API + separate inference service (multi-museum, 100+ concurrent)
- Thermal-adaptive capture ladder on glasses (battery/heat)
- Offline narration bundle on glasses for zero-latency common exhibits
v2.0 — Experience (6–12 months)¶
- Indoor positioning / wayfinding overlays ("navigate to the next tour stop")
- Personalization: adaptive tour depth (kids / expert / accessibility modes)
- Live analytics dashboard for curators (heat-maps of attention)
- Voice Q&A: "tell me more about this artist" (on-prem LLM)
v2.x — Platform¶
- Multi-site management console (museum groups / ministries of culture)
- Content marketplace: share/curate exhibit packages across institutions
- Accessibility: sign-language avatar track, audio-description mode for low-vision visitors
4. Differentiators to protect¶
- On-prem data sovereignty — no cloud egress of visitor imagery (procurement-critical in the GCC).
- Dwell "look & listen" UX — no buttons, no numbers, no app.
- Arabic-first multilingual pipeline with human approval — built for the region, not translated into it.
- Minutes-to-enrol content tooling — museums stay self-sufficient.