Proof-of-Concept Playbook¶
A 4-week museum pilot of the AR Smart Glass Guide · HumanityAI'd — July 2026
Confidential
Download
1. Purpose¶
This playbook lets a museum run a low-risk, time-boxed pilot of the AR Smart Glass Guide in one gallery, and lets both sides judge success against agreed numbers. It is deliberately small: one gallery, ~25 exhibits, a handful of glasses, four weeks.
2. What the pilot proves¶
- Visitors can put on glasses and hear the right narration in their language, hands-free, in under ~1.5 seconds — with no app to download.
- The museum's own content team can enrol exhibits and generate multilingual narration in-house in minutes.
- The system runs entirely on a server inside the museum — no visitor data leaves the building.
- Engagement can be measured (dwell time, play-through, tour completion).
3. Scope¶
| In scope | Out of scope (pilot) |
|---|---|
| 1 gallery, 20–30 exhibits | Whole-museum rollout |
| 8 languages | Custom language beyond the 8 |
| 5–10 glasses units | Large rental fleet logistics |
| On-prem GPU server (loaned) | Permanent server procurement |
| Engagement analytics dashboard | Integration with ticketing/CRM |
4. Hardware & environment checklist¶
- [ ] AR glasses: 5–10 × RayNeo X3 Pro (HumanityAI'd supplies for the pilot)
- [ ] AI server: 1 × GPU workstation/server, NVIDIA GPU 16 GB recommended (8 GB works for recognition-only serving; 16 GB gives head-room for on-box narration generation). 32 GB RAM, 8 cores, 512 GB SSD.
- [ ] Network: dedicated Wi-Fi SSID in the pilot gallery, ≥ 2 access points, wired uplink to the server; server on a static LAN IP.
- [ ] Charging/hygiene: charging station + wipes at the issue desk.
- [ ] Power & rack space for the server near the gallery or in the comms room.
5. Timeline¶
| Week | Activities | Owner |
|---|---|---|
| 0 — Prep | Kick-off, success metrics signed, gallery + exhibit list chosen, site survey (Wi-Fi heat-map), server delivered & installed | Both |
| 1 — Content | Photograph the 25 exhibits (1–3 angles each), enrol them, author/generate the 8-language text + narration, curator review & approval | Museum content team + HumanityAI'd |
| 2 — Trial run | Staff induction (1 hr), internal walkthrough, tune recognition thresholds, fix content gaps | Both |
| 3 — Live pilot | Real visitors with glasses at the issue desk; daily analytics review; collect visitor feedback | Museum ops |
| 4 — Evaluate | Measure against KPIs, visitor survey, readout workshop, go/no-go for rollout | Both |
6. Enrolment procedure (per exhibit, ~5–10 min)¶
- Photograph the exhibit from 1–3 angles (well-lit, framed as a visitor would see it).
- In the admin panel: create the exhibit (title, artist, period, category, inventory no.).
- Upload the photos — the system builds the visual index automatically (no restart, no training).
- Write or auto-draft the short/full description, fun fact, and narration script.
- Generate narration for all 8 languages; curator reviews and approves each clip.
- Done — the exhibit is live for the glasses.
Reference point: the internal pilot enrols a new exhibit in under 5 minutes including 8-language narration.
7. Success metrics (agree & sign at kick-off)¶
| KPI | Target | How measured |
|---|---|---|
| Recognition accuracy on enrolled exhibits | ≥ 95% correct | Staff test pass + visitor reports |
| Time from gaze to narration | ≤ 1.5 s (p95) | System metrics |
| Visitors completing ≥ 5 narrations | ≥ 60% of equipped visitors | Session analytics |
| Dwell-time uplift on narrated exhibits | +30% vs. baseline | Analytics vs. pre-pilot baseline |
| Content enrolment by museum staff, unaided | ≤ 10 min/exhibit | Timed during Week 1 |
| Visitor experience rating | ≥ 4/5 | Exit survey |
| Data egress of visitor imagery | Zero | Network audit |
8. Roles¶
- Museum: gallery & exhibit selection, content sign-off, Wi-Fi & power, front-desk staffing, visitor recruitment.
- HumanityAI'd: server & glasses, installation, staff training, content-tooling support, analytics, evaluation readout.
9. Risks & mitigations¶
| Risk | Mitigation |
|---|---|
| Weak gallery Wi-Fi | Site survey in Week 0; dedicated SSID + extra AP |
| Content bottleneck | HumanityAI'd co-authors first 25 exhibits with the curator |
| Glasses comfort/fatigue | Short average sessions; comfort survey; spare units |
| GPU capacity for narration generation | Generate narration off-peak; 16 GB GPU recommended |
10. Exit & conversion¶
At the Week-4 readout: KPI scorecard, visitor survey summary, curator feedback, and a rollout proposal (gallery-by-gallery plan, fleet sizing, licence & support quote). Pilot content and analytics carry forward into production — no rework.