Building a machine learning platform to scale pet monitoring in real-time

The story

Imagine you take a picture of your pet to diagnose their health, and instantly know what to do. This is the DIG Labs promise. We built a visual AI engine, trained on hundreds of thousands of images - expert reviewed - that can analyze pet issues in real-time across dozens of areas of concern like ears, nose, teeth, and more. But it doesn’t stop there - we use these insights to create personalized plans for pets like diet transition, weight loss programs, and more.

Tech stack


AI models


images used in training


submission every week

About DIG Labs

DIG Labs supercharges pet care, nutrition, lab diagnostics, and public health companies with visual insights that go beyond the surface. Their technology enhances product and customer experiences, setting a new standard in the industry.

With over 20,000 analyses and counting, the company is committed to revolutionizing pet health. As the team continues to grow, significant strides are made in helping thousands of dogs achieve better health each month.

DIG Labs is dedicated to creating better solutions for pet care – less guesswork and more answers, resulting in happier, healthier dogs. Whether it's pet nutrition, diagnostics, or collaborating with leading veterinarians, their focus is on pioneering pet health through data excellence.

The challenge

A complex dataset and data model. To conceptualize the platform and Machine Learning workflow across multiple pet types, diverse pet samples, health indicators, and the company’s proprietary unique classification system that carried its own complexity.

Technology infrastructure for scale. To engineer a a state of the art Machine Learning pipeline that would enable the platform to operate with tend of thousands of requests per day, and connect seamlessly with a multi-tenancy platform. Each tenant (aka Client) may have custom requests and potentially need its own proprietary model.

Product & process

To meet business needs, we need to connect the Machine Learning pipeline to a multi-tenancy product that could scale to the custom needs of enterprise clients easily. The product is divided in three components:

1. ML Pipeline

Run, monitor, and process images to get instant health reports on your pets.

Skip to section

2. Multi-tenancy platform

Onboard new clients, allow us build custom feature sets, and for clients to interface with.

Skip to section

3. API-first product

For innovators in the Health, Pet, and Vet industry to build on top of the ML stack.

Skip to section


Machine Learning pipeline

We see the Machine Learning pipeline as its own application, separated from the multi-tenancy platform. By design, this enables the teams to move faster, isolate problems, and allow for greater f lexibility in development.

The end to end workflow from data capture to pre-processing, model selection, tuning, to results, needed to operate in real-time, ensure transparency in the system - allow us to intervene when we needed to, while incorporating human feedback to improve model performance over time.

Building a client interface

The platform caters to a diverse range of clients, including individual pet owners, vets, each with unique access and data requirements. We enabled this through our multi-tenant architecture, allowing for distinct data streams and interactions for different user groups. The iOS App and web application serve as the primary interfaces for our clients, providing a seamless and intuitive experience. These applications connect to our system through an API Gateway, which manages and directs requests efficiently, ensuring a responsive and secure user experience.

Data engineering

Our core data engineering strategy integrates several data streams into a cohesive framework. We’ve employed AWS Athena for its powerful data query capabilities in order to effectively manage and analyze large datasets. PostgreSQL operations help us manage our Metadata while Amazon S3 supports image storage and management, providing scalable and secure solutions for our extensive datasets. All these components function in parallel to create a robust data warehouse that underpins our ML processes.

Machine learning workflow orchestration

Leveraging SageMaker Studio, we've desgined an efficient machine learning workflow, incorporating data preprocessing, model training, and hyperparameter tuning into scalable SageMaker pipelines. This integration facilitates rapid development and deployment, streamlining our ability to tailor and scale solutions across various tenants efficiently.

Our system's modular design ensures swift adaptation to different pet health concerns, significantly reducing the time required for scaling new tenant-specific models by 5x. This efficiency in scaling and retraining models enhances our predictive analytics, improving decision-making for veterinarians and pet owners.

Sagemaker pipeline

The entire machine learning workflow was orchestrated using Sagemaker pipelines that helped in integrating Continuous Training (CT) ensuring models are accustomed to any new data or pattern variations. What does this mean? SageMaker ProcessingJobs is used for pre-processing and merging data, setting the stage for effective model training. Whereas the SageMaker TrainingJobs handles the training itself, including the crucial step of hyper parameter tuning at scale. This systematic approach ensures our models are optimized for performance and making them highly effective in analyzing pet health indicators from various inputs.

Model registry and expert reviews

After models are trained through our machine learning workflow, they are registered in a Model Registry, where they undergo A/B testing to validate their effectiveness. This stage is critical and involves experts to review models for accuracy, performance, and reliability. The expert review includes a thorough examination of model cards, documenting the performance metrics and use cases of each model. This rigorous process ensures that only the most effective and reliable models are approved for deployment.

Model deployment

We’ve strategically deployed our model through a RESTful API, enabling easy and flexible integration into our client-facing applications. This API serves as the conduit through which our trained and validated models are accessed, providing real-time pet health diagnostics based on the inputs received from the iOS App or web application. The deployment process is designed to be seamless and efficient, ensuring that our clients receive instant, accurate insights into their pet's health.


Multi-tenancy platform

This platform consumes the Machine Learning APIs at scale to build real business use cases. We are essentially the first consumers of our own AI product. Tenants also known as Clients, identified with unique identifier would have a custom version of the application to meet their needs - data requirements, model configuration settings, unique consumer workflows, etc.

A multi-tenancy architecture helped us secure each environment for enterprise scale. In addition, we built a turnkey app that any customer can whitelabel, deployed within seconds.

Decreased client onboarding from

12 hours to under 10 minutes

Automation of Tenants

We automated the creation of tenants to help onboard customers faster. From 12+ hours to configure a tenant to under 10 minutes. This encompassed everything from tenant basic settings, like their name, colours, brand, logo, to deployment on Github and AWS instantly. We built a small app using Retool for the DIG team to manage the process.

Integrations and system openness

While the Machine Learning pipeline was concerned about model performance, what we did with those insights was entirely up to us. We integrated with many third party tools, offering them to Clients who wanted to integrate with the wider ecosystem. We were able to send insights to Practice Management Softwares like Shepherd, engagement tools like Segment or Airtable, or communication tools like Gladly.


API-First product

DIG Labs offers multiple products and solutions to Clients. The element of customizability, personalization, and control for enterprise use cases, required us to build a product that is extremely adapatable to current known use cases but also the future.

Exposing our APIs to clients to build on top fulfilled the promise of an open-architecture system. Enterprises across Pet, Vet, and Heath industries built on top of our RESTful API to create custom experiences inside their own applications. This opens the door for dozens and sometimes hundreds of use cases.

Meet our open API stack

Externalizing your APIs is one of the easiest things to do to increase revenue and create new profit centres.

Swagger and Postman for API documentation. We automated documentation with YAML files so it’s effortless to maintain.

AWS API Gateway to manage credentials, rate limits. It also allows you to offer an SDK in almost any language.

Loom for technical demos and walkthroughs.


Our solutions helped DIG Labs and its clients achieve unparalleled accuracy of pet image analysis due to the compartmentalization of the Machine Learning pipeline and its robust feedback system. We automated new Client on-boarding by creating a multi-tenancy platform, built a system that was able to meet product customization requests - the choice between white label or a custom solution very quickly.

From infrastructure, to the core application, an open architecture allowed us to iterate quickly, build for speed, and pivot with the business when needed. We create a platform that is consumed in many different ways including an API-first product, with little engineering effort along the way to broaden our market reach and diversify the use cases.


AI model accuracy


Expert-labelled images used for training


Machine Learning models deployed


Human speed with image analysis and recognition


image submissions processed per week

Island’s capability to source and retain top talent is unmatched. I have full confidence that his team will deliver, and is scalable to our growing needs.

Tara Zedayko
CEO at DIG Labs

Interested in learning more?

Opening Motion Calendar...

Congrats, you may be CDAP grant eligible!
Book a consultation with our founder on another tab.
Facing issues? Click here to try again.

Oops! Something went wrong while submitting the form.