Two weeks ago I wrote about building an offline-first livestock counter with YOLOv8 and CoreML. Today it's a real product on the App Store.
HerdCount — Count your flock, even offline
£3.99. No subscription. No cloud. No account. Pay once, use forever.
What It Does
Point your phone at livestock or plants. Tap a button. Get the count.
HerdCount uses on-device AI (YOLOv8 + CoreML) to detect and count chickens, sheep, cattle, and plants from a single photo — in under a second, with zero internet required.
Why I Built It
I work with on-device computer vision professionally — building Axsy Smart Vision, an AI-powered field inspection platform for Salesforce. Retail planogram detection, product identification, compliance scoring — all running on-device.
But the agricultural space has a simpler, more immediate problem: counting animals is tedious and error-prone. Farmers do it by eye, multiple times a day. Miss one sheep and you're searching hedgerows at dusk.
The same on-device ML pipeline I use for retail product detection works beautifully for livestock. So I built it.
The Technical Stack
- YOLOv8 — trained on livestock datasets, converted to CoreML
- On-device inference — runs on the iPhone's Neural Engine, no cloud round-trip
- Offline-first — works in fields, barns, anywhere with no signal
- Swift/SwiftUI — native iOS, 8.7 MB total
- Export — CSV via AirDrop, email, or Files app
The model handles overlapping animals (tuned IoU threshold at 0.3 rather than the default 0.5) and lets you tap false positives to remove them. Manual +/− adjustment before saving.
What I Learned Shipping It
1. App Review is opinionated. Apple rejected the first submission because the category detection UI wasn't clear enough. Fair feedback — I redesigned it and it's better for it.
2. The model is the easy part. Training YOLOv8 and converting to CoreML took a weekend. The other 90% was UI polish, edge cases, CSV export formatting, and App Store screenshots.
3. Pricing matters. I went with £3.99 one-time purchase. No subscription, no ads, no data collection. Farmers are practical people — they'll pay for a tool that works but won't tolerate dark patterns.
4. On-device AI is a real differentiator. Every competing app I found requires internet. That's a non-starter for someone standing in a field in rural Wales.
From Blog to Product
The Dev.to post about the technical approach got genuine engagement — @gimi5555 asked about NMS strategies for clustered animals, which led to a good discussion about density estimation as a fallback.
That conversation validated the approach. Two weeks later, it's a shipped product.
If you're working with on-device ML and sitting on something useful — ship it. The App Store review process is less scary than it looks, and real users find real problems you'd never catch in development.
Built by RT Sullivan Consulting. I write about on-device AI, Salesforce field apps, and shipping real products at dev.to/toddsullivan.
Top comments (0)