Home / Services / Computer Vision
Capability

See what your cameras already see.

Vision AI for the physical world — OCR, document intelligence, quality control, perimeter monitoring, ANPR, and count/inventory. Trained and tuned for real warehouses, sites, branches, and back offices in East Africa.

Edge or cloudPrivacy-first options
RTSP-nativeYour existing cameras
PDPA-compliantNo faces stored by default
Build via Integration Send a sample
Architecture
Edge-first. Private by default.
Camera / device RTSP · USB · IP Smartphone Edge node Jetson · Intel NUC Runs models locally Detect + track YOLO · Segment OCR · ANPR Rules + VLM Zone · time · logic Anomaly description Event Log · alert Webhook · SMS Only metadata leaves the site by default · video stays local unless required Active-learning loop · mislabelled frames re-fed into training
What changes
Physical ops, before and after.
Before

Cameras that only matter after an incident.

  • Hours of CCTV scrolled through to investigate a single event.
  • Gate-log books filled by tired guards with inconsistent handwriting.
  • Paper waybills and invoices re-keyed into ERP — with errors.
  • Quality defects caught at customer, not at production line.
  • No reliable count of what actually entered or left the premises.
After

Cameras as operational sensors.

  • Every vehicle logged by plate, timestamped, cross-checked to manifest.
  • OCR turns documents into structured data in seconds.
  • Defects flagged on the line with photo evidence attached.
  • Daily dashboards: counts, dwell times, zone occupancy, exceptions.
  • All video stays on-site unless specifically exported for a case.
Typical stack
Proven, not exotic.
YOLO v8/v10
SAM 2
PaddleOCR
ANPR kit
Jetson
ONNX
DeepStream
OpenCV
Gemini VLM
Label Studio
MLflow
MinIO
How we deliver this
The same capability, three ways.
In an Audit

Vision feasibility

Site visit, camera inventory, sample footage analysis, and a recommendation: which use cases are ready, which need new cameras or lighting, and the ROI math for each.

In an Integration

One vision system in production

Typical: ANPR at gate, perimeter anomaly, or inline document OCR. Edge-deployed. Event stream into your ops dashboard. 10–14 weeks including annotation & tuning.

In a Transformation

Site-wide vision fabric

Multiple cameras across multiple sites feeding a unified ops view. Rules engine shared across use cases. Retraining pipeline with analyst feedback. Deployable to new sites in days.

In production
A gate event, in under a second.
ops.vision — Gate A · live
Frame captured
vehicle 0.96
Cam-03 · 02:14:07 EAT
ANPR
KBZ 481T
confidence 0.98 · 142 ms
Match
Manifest MAN-9918 · expected 02:10–02:30
Gate opened · event posted to WMS
Data retention
Metadata 90d · Frame 7d · Video 0d (unless flagged)
Next step
Send us 60 seconds of footage.
We'll run a no-cost feasibility pass: tell you what's detectable with current cameras, what needs upgrading, and what the ROI math looks like.
Start with an Audit Send a sample