Barkyn AI Health
Challenge
Barkyn is a Portuguese pet-care company whose mission is preventive dog health. Most owners only learn about a health problem when symptoms are already visible, Barkyn wanted to move that moment earlier, with an app that turns a photo of a dog into actionable health guidance.
That framing created three hard engineering problems. First, the product is camera-first: the entire experience depends on capturing consistently good photos of an uncooperative subject, a dog, across a huge range of devices, lighting conditions, and camera qualities, while staying fast and intuitive enough for everyday use.
Second, the intelligence had to run on the device. Real-time dog detection and pose guidance during capture can't wait for a network round trip; the app needs to recognise the dog and guide the owner into a usable shot as they frame it, which means machine learning running locally, within the performance and battery constraints of ordinary phones.
Third, on-device AI capabilities were evolving quickly and remained partly experimental. The team had to keep integrating a rapidly changing ML component without compromising the stability of the app or the release cadence, shipping continuously while the most novel part of the stack kept moving underneath.
Barkyn is a Portuguese pet-care company whose mission is preventive dog health. Most owners only learn about a health problem when symptoms are already visible, Barkyn wanted to move that moment earlier, with an app that turns a photo of a dog into actionable health guidance.
That framing created three hard engineering problems. First, the product is camera-first: the entire experience depends on capturing consistently good photos of an uncooperative subject, a dog, across a huge range of devices, lighting conditions, and camera qualities, while staying fast and intuitive enough for everyday use.
Second, the intelligence had to run on the device. Real-time dog detection and pose guidance during capture can't wait for a network round trip; the app needs to recognise the dog and guide the owner into a usable shot as they frame it, which means machine learning running locally, within the performance and battery constraints of ordinary phones.
Third, on-device AI capabilities were evolving quickly and remained partly experimental. The team had to keep integrating a rapidly changing ML component without compromising the stability of the app or the release cadence, shipping continuously while the most novel part of the stack kept moving underneath.
Solution
Pink Room delivered the engagement as a combination of AI consulting, software engineering, and product and project management, shaping the product as well as building it.
The app was built in React Native (with React Query), delivered through Expo, with Clerk handling authentication and Intercom for in-app support and messaging.
The core of the build is the camera-first flow backed by on-device machine learning. A local ML model performs real-time dog detection and pose identification while the owner frames the photo, guiding them toward a capture that's good enough for accurate analysis, solving the image-quality problem at the moment of capture rather than trying to repair bad photos afterwards. Because the model runs on the device, the guidance is instant and the experience works reliably in real-world conditions.
From the captured photos, the app analyses physical indicators: fur, eyes, teeth, and produces a health score from 0 to 100, with personalised recommendations that turn the score into concrete preventive-care actions.
Around the ML core, the engineering discipline was about containment: keeping an experimental, fast-evolving on-device AI component integrated cleanly so the app around it stayed stable and releases kept shipping on cadence.
Pink Room delivered the engagement as a combination of AI consulting, software engineering, and product and project management, shaping the product as well as building it.
The app was built in React Native (with React Query), delivered through Expo, with Clerk handling authentication and Intercom for in-app support and messaging.
The core of the build is the camera-first flow backed by on-device machine learning. A local ML model performs real-time dog detection and pose identification while the owner frames the photo, guiding them toward a capture that's good enough for accurate analysis, solving the image-quality problem at the moment of capture rather than trying to repair bad photos afterwards. Because the model runs on the device, the guidance is instant and the experience works reliably in real-world conditions.
From the captured photos, the app analyses physical indicators: fur, eyes, teeth, and produces a health score from 0 to 100, with personalised recommendations that turn the score into concrete preventive-care actions.
Around the ML core, the engineering discipline was about containment: keeping an experimental, fast-evolving on-device AI component integrated cleanly so the app around it stayed stable and releases kept shipping on cadence.
Results
Barkyn AI Health shipped as a working consumer product: dog owners photograph their pet and receive an on-device AI health score (0–100) with personalised preventive-care recommendations, turning everyday signals into actionable guidance, all within a single mobile app.
The on-device approach proved itself in production. Real-time detection and pose guidance made the camera experience reliable across the diverse devices and conditions of a real consumer user base, and running the intelligence locally kept the experience fast and private; no photo has to leave the phone for the guidance to work.
For Pink Room, the project is our clearest shipped example of on-device AI in a consumer app: a machine-learning model running on the phone, wrapped in product engineering that made an experimental capability dependable enough to build a product on. The engagement ran from June 2024 to May 2026, spanning product strategy, engineering, and project management, and ended with a clean handover to Barkyn's in-house team, including onboarding their new engineer, leaving the product and its codebase in a state the client could carry forward independently.
Barkyn AI Health shipped as a working consumer product: dog owners photograph their pet and receive an on-device AI health score (0–100) with personalised preventive-care recommendations, turning everyday signals into actionable guidance, all within a single mobile app.
The on-device approach proved itself in production. Real-time detection and pose guidance made the camera experience reliable across the diverse devices and conditions of a real consumer user base, and running the intelligence locally kept the experience fast and private; no photo has to leave the phone for the guidance to work.
For Pink Room, the project is our clearest shipped example of on-device AI in a consumer app: a machine-learning model running on the phone, wrapped in product engineering that made an experimental capability dependable enough to build a product on. The engagement ran from June 2024 to May 2026, spanning product strategy, engineering, and project management, and ended with a clean handover to Barkyn's in-house team, including onboarding their new engineer, leaving the product and its codebase in a state the client could carry forward independently.
