Sound Mathematics is an innovative UK startup that specialises in non-destructive testing. Their research team has won numerous grants and awards for their core technology: a Machine Learning algorithm producing full health reports for metallic components. When we met their team the first time, they were at the late stage of building their first MVP (or minimum viable product). However, they were not set up to deliver their insights as a service. They needed robust and cost efficient software architecture to start serving their first clients. We decided to partner with them to build and host all their inference pipelines on our cloud.

Challenges

After compiling a full list of requirements for the architecture, we highlighted the key ones to guide our decision making:

• Users of the service are required to upload large files.
• We return a complete analysis in a matter of seconds.
• Preferably the cost at rest should be as close as possible to zero without impacting the potential to scale.
• Project duration: less than a month to avoid impacting commercial partners.

Implementation

Serverless

We designed and built a serverless architecture to store, process, and return predictions on a

Containerisation

All the software we

Multistage Processing

Users are expected to upload large files to our backend together with special configurations. This process can be slow for the user and costly for us. We split the inference process atomically to catch user errors as quickly as possible and allow users to break up the data upload and result collection process.

Results

Thanks to the joint effort of our teams, Sound Mathematics was able to quickly integrate the APIs we built for them in their frontend and cloud architecture to engage with their customers and keep moving forward with commercialisation.