Rebuilding a music education platform on Kubernetes so lessons scale on demand
MuseCool is a London-based music school providing one-to-one instrumental lessons for children, available in-person and online, matching students with vetted tutors and supporting home practice through their proprietary lesson-understanding technology, The Muse.
There is no question that Digital Back Office's depth of expertise is what got us ready for our next round of funding. From AI to DevOps, they understood exactly what we needed, and brought real cost optimization into both. I honestly can't believe anyone else could have done this for us. What stood out most was how well they understood startup mindset. Everything moved fast, and delivery was always quick.
Problem
MuseCool's platform was built on a monolithic architecture running on VMs, unable to handle more than two parallel music lessons at once. Processing class recordings placed heavy load on the GPUs, and the VMs would crash under that load, capping how many lessons could run at the same time, right as the company was heading into its next funding round.
Solution
We rebuilt MuseCool's cloud infrastructure around independent microservices in Kubernetes, separating GPU-heavy workloads from the rest of the system so they could scale independently as class volume grew, and migrated the platform end to end from MLOps through to the UI.
Results
MuseCool can now onboard users without worrying about platform scalability, and can scale their AI model with confidence.
The platform's monolithic architecture meant every class recording competed for the same limited GPU resources, and there was no way to isolate or scale that workload independently, so growth in one area risked taking down the whole system.
Running GPU capacity around the clock, regardless of actual demand, meant infrastructure costs would only grow as usage grew, with no mechanism to bring them back down during quiet periods.
We assessed the existing monolithic setup to identify exactly where the bottlenecks were, and mapped out which workloads needed to be isolated and independently scaled.
We decomposed the platform into independent services in Kubernetes, separating GPU-heavy recording processing from the rest of the application, and connected workloads through message brokers (Google Pub/Sub, AWS SQS).
We moved infrastructure management to code using Terraform and Helm Charts, and stood up separate staging and production environments to support safe, repeatable deployments.
With the architecture in place, we built CI/CD pipelines to manage the application lifecycle end to end, and continued ongoing feature development.
Kubernetes event-driven auto scaling brings GPU capacity online only when classes are actually running, and scales it back to zero when demand drops, so MuseCool never pays for idle GPU infrastructure.
The microservice architecture is built to recover from failure on its own, keeping the platform reliable without requiring manual intervention every time something goes wrong.
CI/CD pipelines automate the path from code to production, letting the MuseCool team ship new features faster and with more confidence.
From AI strategy and automation to custom machine learning applications, we create solutions that help organizations innovate faster, operate smarter, and scale with confidence.
Complete the form below, and we'll respond within 24 hours.
We are based in UK, but we can support your business in any time zone, we work 24x7.
Difficult is done at once, the impossible takes a bit longer. If it is possible, consider it done. The impossible, it will be done!
We provide full spectrum of enterprise data services.
We understand that every business is different. We work on both fixed cost and time and material basis. We can also provide consultants, who can work alongside your existing team.
