Smart hardware to the dog's rescue
Aniv8 is a biotechnology startup based in the United States, founded by accomplished veterinary medicine experts. Together with our team of AI and hardware consultants, they created an innovative, first-ever wearable sensor for dogs
Their idea is to detect and diagnose osteoarthritis based on animal movements automatically. It's is a painful condition, and it affects about a quarter of all dogs.
The project was designed to be a subscription-based service, initially targeted at professional veterinary clinics. At first, the startup wanted to test the smart collar only on dogs. At a later stage, the idea was to broaden the scope to other animals.
The challenges of an innovative idea
Aniv8 needed a partner who could help them create a viable end-to-end solution. This involved designing the software platform, creating hardware electronics and sensors, and delivering software that could read the hardware data with Bluetooth Low Energy. Finally, to develop mobile and desktop applications for the end users and vets.
After acquiring funding, the startup started searching for a technology partner who could help them find and implement the most suitable software and hardware solution. At first, they wanted to test an MVP – to check if it’s even possible to use machine learning to solve such a challenge. Having spent over 180 000 USD on collaboration with other vendors, the search didn’t bring the expected results. The budget was shrinking.
Punktum's technological expertise saves the day
When the company reached out to us, their solution was still at a very low Technology Readiness Level (so-called TRL), which meant it was still uncertain if the solution would ever work, and how to prove that. They needed a solid methodology to support further progress in the project.
Acting as an Interim Product Owner, we conducted discovery meetings and workshops with the Aniv8 team. Our experts analyzed the current state of affairs and proposed a new roadmap to help move forward swiftly
Deep learning - the optimal solution for this goal
We proposed a deep-learning approach to make the initial classification, assess the severity of the illness and then could be used to extract data and monitor the disease further. We have decided to employ deep learning models and a deep convolutional neural network for imaging data to recognize patterns based on collected raw data.
Before developing the algorithms, our team – which included international scientific PhDs – quickly identified that what was first needed is to clean the learning data (dogs’ activities) to ensure it was correctly annotated and workable.
Using that data, our experts developed the Machine Learning algorithm and product concept, which verified positively for the client that the solution could be built and scaled up. Our overall solution involved the development of the right processing AI algorithms and models of the data from sensors (accelerometers and gyroscopes).
The results of our efforts allowed us to sort through all the data and extract the deep features to classify OA severity for a given dog. Next, we defined milestones and possible further steps on the road to creating the product the startup envisioned.
Our partner was delighted with the results of the collaboration. Especially valuable was our speed of delivery and the quality of the smart solutions we have provided on a fixed, tight budget for a product with a still very low technology readiness level, in the prototype phase.
This solution in numbers
Our process: from rapid project initiation to positive prototype test and delivery
Discovery and analysis with the client to understand their business context and high-level goals.
Auditing the work conducted by the previous partners
Assessing the tech feasibility of the product
Planning the innovation process milestones and a product roadmap
Proposing a development methodology and a solution concept to create a viable product
Algorithm development and verification
Drawing the technology blueprint for further, larger-scale tests
Steps to achieve the optimal solution
Cleaning the client’s data: our experts started here as this step was necessary before proceeding with further assessment and working on an optimal solution.
Proving the product feasibility: the goal was to establish if the product is feasible and if the further work on the deep learning model was possible and achievable in further development.
Proposing a solid methodology to increase TRL: after conducting thorough tests, the product was determined feasible, so the naturally crucial next step was to establish the methodology for further development, learning models and possible broadening of the scope onto other animals, which were possible and achievable at the time.