All apps
AI-native

Visionline

Computer-vision analytics for retail foot traffic and dwell-time heatmaps

Visionline is the AppLiaison answer for operators who want to know what is actually happening on their physical premises — foot traffic counts, dwell-time heatmaps, queue depth, conversion of walk-in to checkout. It ingests video from existing IP cameras, runs detection inference (cloud or edge), and renders an operator dashboard that shows the metrics ops actually steers on, not raw motion blobs. It carries forward the VisionAI listing from the legacy AppLiaison site.

Key features

  • Person detection and tracking via YOLOv8 / RT-DETR (swappable models)
  • Foot-traffic counts (entry, exit, by-direction, by-zone)
  • Dwell-time heatmaps overlaid on customer-supplied floor plans
  • Queue-depth measurement with configurable alert thresholds
  • Conversion measurement (walk-in to checkout) with retail POS integration
  • On-device or cloud inference per camera (latency / cost tradeoff per site)
  • Alerts to email / SMS / Slack on threshold breaches
  • Reporting (hourly / daily / cohort) with CSV + BI-tool export

Architecture

Architecture variant: ai-native
Frontend
Next.js 14 (operator dashboard)WebSocket live tiles (current count, current dwell)Recharts for analytics
Backend
FastAPI (Python 3.12) for ingest + inferencePyTorch + torchvision (model serving)YOLOv8 / RT-DETR detectors (swappable)
Data + infra
Vercel (dashboard)Modal or Lambda Labs for cloud GPU inferenceOn-device runtime via ONNX or CoreML / TFLite
Integrations
RTSP / ONVIF cameras (direct ingest)Existing NVR/VMS (Milestone, Genetec) via ONVIF passthroughTwilio (alert SMS)Slack and Microsoft Teams (alert routing)

Built on the Computer Vision Stack.

What you get

Visionline lands as an operator dashboard under your brand and an inference layer that connects to the customer’s existing IP cameras. Per-site, the integration team decides whether inference runs in the cloud (lower latency budget) or on an edge node (NVIDIA Jetson, Mac mini, or Intel NUC with OpenVINO) for sites that need it.

Sample customer story (placeholder — illustrative only)

A 60-location regional convenience-store chain licensed Visionline to escape a RetailNext install whose per-store fee had climbed past their ROI threshold. The win after one quarter was the queue-depth alerting: store managers got a Slack ping when checkout queues exceeded 3 customers, and the chain’s average customer-perceived wait time dropped 18% in their post-deployment NPS survey.

What’s NOT included

  • Camera hardware sourcing or installation. Customer brings their own; we partner with installers as introductions.
  • Custom training of customer-private models. A paid engagement with a separate timeline.
  • Long-term video archival beyond 90 days. S3-IA bucket lifecycle is configurable, customer-paid.
  • Real-time human review of every event. The system is autonomous; alerts go to humans, not every detection.
  • Loss-prevention / theft detection. Possible as a paid Pass-2 customization with the right dataset; not a default.

Whitelabel surfaces

SurfaceThemeableCustom domain
Operator dashboardyesyes
Email + SMS alertsyesyes (email sender domain)
PDF reportsyesyes
REST API + webhookyesyes