⌘K
ScopePeriod

Safety AI Monitoring Center

Computer-vision safety analytics · supervisor-in-the-loopAll operations
Human review mandatory — no automated employment decisions; analytics are aggregate decision support. Faces are blurred at the edge device, events are reviewed by a supervisor before any follow-up, and detections are never used for disciplinary automation. Footage retention 30 days (synthetic policy).
Cameras online
46/48
Edge inference · 1 offline
Events (7 days)
6
4 reviewed · 1 false positive
Awaiting review
2
Target: < 4 h to review
PPE compliance
96.4%
+1.2 pp vs last month
Days since LTI
12
Group · last: Sep 16
Predicted near-miss (Oct)
28
+12% vs Sep · AI forecast

Live camera wall

6 of 6 priority feeds in scope · simulated stream

Violation Proximity Compliant Faces blurred at edge · silhouettes onlySimulated footage

Leading-indicator heatmap

Risk index by zone × hour (PPE lapses, proximity, congestion, heat stress) · click a cell

06:0008:0010:0012:0014:0016:0018:0020:00
Mill A · Loading ramp4444385669908054
Mill A · Boiler area4345486980655041
Mill A · Press station2838516151322423
Mill B · Sterilizer deck4438667162513627
Mill B · Workshop2233373256786135
Estate C · CP-43235346477564040
Estate D · CP-21927505850351825
Low → high risk index (0–100)14-day rolling window · synthetic

Safety event review queue

2 awaiting supervisor review

Safety Vision v1.4
14:32Mill A — Boiler Area27 Sep
PPE compliance warning · Person detected without required helmet
Status: Supervisor Review RequiredConfidence93%
11:05Mill A — Loading Ramp27 Sep
Vehicle–pedestrian proximity · Pedestrian within 2.1 m of reversing truck TR-006
Status: Reviewed — Coaching LoggedConfidence88%
09:47Mill B — Sterilizer Deck27 Sep
Restricted area · Entry to sterilizer pit zone during cycle without permit tag
Status: Supervisor Review RequiredConfidence81%
16:20Mill B — Workshop26 Sep
Unsafe behaviour · Grinding without face shield
Status: Reviewed — ConfirmedConfidence77%
08:12Mill A — Press Station26 Sep
PPE compliance warning · Ear protection not detected in >85 dB zone
Status: Reviewed — False PositiveConfidence84%
13:55Estate C — Collection Point CP-425 Sep
Vehicle–pedestrian proximity · Harvester crossing behind loading tractor TC-03
Status: Reviewed — Coaching LoggedConfidence79%

Incident trend & prediction

Monthly near-miss, first-aid and LTI · Oct = AI forecast

Incident prediction — next 30 days

Leading indicators → incident likelihood

Prediction
Near-miss (Oct)
28
vs 25 in Sep
P(recordable)
34%
group, 30 d
Hotspot
Mill A ramp
12:00–18:00
Drivers of predicted near-miss increase
Ramp congestion (FFB queue ↑)
+6.8 pp
Peak-crop season volume
+4.9 pp
Mill A maintenance overtime
+3.6 pp
ST-04 bearing job (LOTO)
+2.4 pp
PPE compliance trend
−1.8 pp
◀ lowers predictionraises ▶
Recommended controls
Confidence76%
Deploy a ramp spotter 12:00–18:00 at Mill A, pre-job briefing & LOTO permit for the ST-04 bearing replacement, and fatigue check for technicians above 24 h overtime.

Privacy, governance & model card

  • • Faces are blurred on the edge device; only anonymised silhouettes and event crops leave site.
  • • Every event is reviewed by a supervisor before any follow-up — confirm, false positive or coaching.
  • • Detections are not used for disciplinary automation or individual performance scoring.
  • • Retention: event clips 30 days, aggregates 24 months (synthetic policy).
  • • Access: HSE Manager & site supervisors; audit trail on every review.
  • • False-positive feedback is used to retrain the model quarterly.
Safety Vision (PPE / Proximity)
Object detection + pose + zone rules · v1.4.0 · Pilot
F1 0.86 · Annotated CCTV frames (consented, anonymised)
Prototype model metrics — demonstration only.
V-TEKI — IT & Business ConsultingDeveloped by V-TEKI · IT & Business ConsultingPrime Agri AI Intelligence Platform — prototype. All operational records are synthetic; model outputs are simulated; financial impacts are scenario estimates; recommendations require human validation. No actual company confidential data is used.