A technical team needing to get to market quickly despite their specialized domain
Audere is a digital health nonprofit developing software to improve global health in the world’s most underserved communities. Their team of passionate, innovative minds combines smartphone technology, computer vision & machine learning, and the best of cloud-based services to deliver healthcare technology solutions worldwide.
Vision AI is a core component of these applications as they upload images scanned in the field to tablets and phones to be analyzed by a neural network. Given the unique domain, the team needed to build a data asset they could leverage to train neural networks for their first key applications evaluating HIV/AIDS, tuberculosis, and later COVID test results. While having a strong technical team with a good understanding of how neural networks work, they struggled to find software that would spare them from the hassle of having to build custom features needed to label the data and train their models.
The challenge was to find tools that made collaboration between the medical and technical teams easier and to save their time to allow them to focus on building the product.
Before discovering Hasty, labeling images was labor-intensive, time-consuming, less accurate, and progression through the groundwork to build our AI detection model was much more frustrating. Hasty's approach of training the model while labeling with faster annotate-test cycles has saved us countless hours. The speed and ease of use have allowed us to accelerate our mission to improve global health in the world's most underserved communities.
Principal Software Engineer, AI (Deep Learning - Computer Vision), Audere
Using AI to train AI in Hasty
Audere switched to Hasty after suffering the pain of having to label everything manually in another platform, despite the promises on the website. “I was super sad when we found Hasty, as I wasted seven months of my life last year in [redacted competing platform].” The automation Hasty provided saved a massive amount of manual work from the first week.
The game-changer here was that the automation came from a model trained on their data. This is opposed to ‘transferring’’ the learning from another domain and pre-label images with an ill-fitting model. With Hasty’s flow, they could automate more and more work as the model trained and retrained, almost in real-time, on their data.
The team could create a significant data asset to build their application for a highly specialized domain without relying on a small army of people to do the manual work. The core team could focus on the remaining tasks to deliver the core application.
Growing the team and use-cases
“Hasty allows AI practitioners to: spin up their own datasets → get image set → move into
Hasty and turn that into models. You don’t have to send it worldwide where maybe quality won’t be that great. Our net turnaround time is fast, and they can verify quality along the
Audere has expanded from one disease to serving applications of multiple illnesses. They have cemented their position in their market and broadened their footprint to give themselves a more substantial base to grow from.