▸_
trioAI Store Manager
LIVE
12:32:00PM EDT
T1$280
T2$105
T3$350#1
T4$0
T5$280
T6$175
CAM 1 · OUTSIDE PATIO
PATIO · TODAY · EST.$1190
PATIO PLAN6 tables
HOSTT1T2T3T4T5T6
DOUBLE-CLICK TO ZOOM
B1$175
B2$280#1
B3$175
B4$105
B5$175
B6$105
CAM 2 · INDOOR BAR
BAR · TODAY · EST.$1015
BAR PLAN6 tables
HOSTB1B2B3B4B5B6
DOUBLE-CLICK TO ZOOM
S1$280#1
S2$105
S3$0
S4$0
CAM 3 · STAGE
STAGE · TODAY · EST.$385
STAGE PLAN4 tables
HOSTS1S2S3S4
DOUBLE-CLICK TO ZOOM
VENUE · LIVE ACROSS 3 CAMS
1 MODEL · 3 CAMERAS
Hog's Breath Saloon · Key West
$2590
today · est. @ $35/cover
30
seatings · 16 zones tracked
PATIO
13
BAR
11
STAGE
6
1 model · 3 cameras · $2.31/day total
trio
Hog's Breath Saloon
3 cameras · Lunch rush · Thursday, June 4
CONNECTING TO 3 CAMERAS
DETECTOR
Watching zones across 3 cams · stride 10s · 30s rolling context window
0/12 EVENTS · 1 MODEL · 3 CAMERAS
Onboarding

From camera URL to live detection · 7 days

Same backbone everywhere; only the small head is fine-tuned per camera. Means month-1 onboarding scales to N stores without re-training the heavy part.

Day 1
Connect
Share your RTSP / NVR / cloud-camera URL. Nothing to install on-site.
YOU
Day 2-3
Zone schema
We draw bboxes on a sample frame; you approve in a web UI in ~30 minutes.
JOINT
Day 3-5
Label + QC
Motion prefilter surfaces ~500 candidates · Qwen3-VL pre-labels them · humans QC ~200 hard cases.
US
Day 5-7
Fine-tune head
Train a ~10M-param head on your data. V-JEPA 2 backbone stays frozen — no expensive retraining.
US
Day 7 → Week 4
Live → demo-grade
Live at ~85% precision / ~5s MAE on day 7. Pilot tuning lifts to ≥95% precision / ≤3s MAE by end of month 1.
LIVE
Daily report · auto-generated

Venue performance · today

Insights derived from 42 seating events detected across 3 cameras since the morning shift began. Revenue figures are model estimates at $35/cover until POS data is connected.

Revenue per table · today

Where the money came from

est. @ $35/cover
Seating intensity by hour

When the patio fills up

illustrative · awaiting historical data
11121314151617181920
T1
T2
T3
T4
T5
T6
B1
B2
B3
B4
B5
B6
S1
S2
S3
S4
0
peak
Cost to operate

Cheap enough to leave running 24/7

cloud pricing 2026-06-01
RECOMMENDED
Shared cloud GPU
$0.77/day
≈ $23/month
H100 amortized across ~30 cameras
Recommended for multi-location franchises
Dedicated cloud GPU
$2.92/day
≈ $87/month
AWS g5.xlarge · 1 NVIDIA A10G per cam
Single-location, no sharing
On-prem (existing hardware)
$0.10/day
≈ $3/month
Mac Studio / RTX 4090 · electricity only
Privacy-sensitive operators
The bottom line
One extra seated cover per month covers the entire system.
Recommended tier costs $23/mo per camera. At $35/cover, break-even is 1 cover/month. Every seating after that is pure margin captured.
Operational observations

Outside patio is the lead earner across 3 cameras

Outside patio captured est. $1488 from 17 seatings; stage captured est. $788 (1.9×). Same backbone runs all 3 cams — only the per-cam head changes.

T3 generated 12% of venue revenue

Round · mid-center (outside patio). 5 seatings observed since the morning shift began · est. $438 contribution at $35/cover.

T4 sat empty for the entire shift

Square · bottom-left — 0 seatings detected today. Worth checking line-of-sight from the host stand or considering a layout adjustment.

3 simultaneous seatings within 6s

B2, T1, S1 all seated in rapid succession during the lunch peak. Server-to-table ratio under pressure — consider floating an extra server during 12:30-1:15 PM.

T2 ran cooler than the venue median

High-top · right — 1 seatings vs median of 3. Possible visibility / seating-flow issue worth observing across the dinner shift.

42 seatings detected today · est. $3675 venue-wide

Across 16 tracked tables on 3 cameras · 12 of these were captured during the live window above. Revenue is a model estimate at $35 per cover until POS data is connected.