HerdLink — the demo, explained
HerdLink is a demonstration of a next-generation analytics layer for virtual-fencing cattle collars: herd-relative behavioural analytics and social-network analysis that flag unwell animals from the telemetry collars already collect — earlier, and with far fewer false alarms, than the per-cow threshold alerts the category ships today.
There is no hardware here. The demo is a fully client-side web app (http://localhost:5173) that simulates a realistic herd, streams synthetic collar telemetry from it, and runs a real detection layer on top. The point it makes: the defensible value in this product category is the ML and UX layer, and that layer can be demonstrated end-to-end in software.
What you're looking at
| Piece | What it does |
|---|---|
| Visual overview | A slide-by-slide walkthrough of the whole idea — start here for the picture |
| Herd simulation | 80 cows with realistic daily behaviour rhythms, herd movement, and social structure |
| Collar telemetry | Synthetic GPS, speed, behaviour, temperature, and rumination streams per cow |
| Health conditions | Injectable ground truth — lameness, illness, oestrus — with realistic behavioural signatures |
| Detection layer | Herd-relative anomaly scoring that sees only telemetry, never the ground truth |
| Social graph | Proximity-derived herd social network — sick cows visibly detach from their cluster |
| Weather & environment | Rain, wind, heat, and day/night, all coupled into behaviour |
| UI guide | What every control, colour, and panel means |
| What makes it different | The three analytics ideas that set HerdLink apart |
The claim
Smarter software, from the same data. Scoring every cow against the herd — and against her own history — rather than against fixed thresholds makes detection robust to weather, season, and feed changes: the whole herd shifts together, the sick individual still stands out. Reading the herd's social network on top of that surfaces withdrawal, one of the earliest sickness signals, with no extra sensor. The detection page explains the method and what makes it different frames the case; the demo lets you verify it against injected ground truth.
Quick demo script
- Open the app, set sim speed to 10 min/s.
- Click Random ill in the sidebar. Note the cow is not visually marked — the ground truth is hidden from the detector.
- Within a sim-hour the cow picks up an amber watch ring; shortly after, a red pulsing alert ring and a plain-language alert in the feed.
- Switch to the Social graph view: the sick cow's edges thin out as it detaches from its grazing cluster.
- Hit Heatwave: the herd migrates to the shade trees, everyone's temperature rises — and no false alerts fire, because detection is herd-relative.