Tuned Global boosts music ingestion tenfold with AWS

Tuned Global has modernised the core infrastructure of its music platform with Amazon Web Services and SourceFuse, increasing music catalogue ingestion capacity tenfold.

The work was designed to prepare the platform for further international growth and heavier use of artificial intelligence across its services. Tuned Global provides licensed music technology and infrastructure to businesses including telecoms groups, fitness companies, media organisations and other brands.

Its platform handles about 200 million music tracks, along with podcasts, audiobooks and video, and supports customers across the US, UK, Europe, Asia Pacific and parts of Africa. As customer volumes grew, some new clients brought catalogues of 30 million to 60 million tracks that had to be ingested, standardised, enriched with metadata and made available through more than 500 APIs.

The project involved moving Tuned Global's core database layer from Microsoft SQL Server to Amazon Aurora PostgreSQL and updating its extract, transform and load architecture with AWS Glue and Amazon EventBridge. It also modernised its master catalogue ingestion database and customer-facing databases at the same time.

Tuned Global acted before its existing systems became a constraint, according to Chief Technology Officer Yahya Bilal.

"We basically sit at the intersection of licence-cleared music and the infrastructure needed to power some of the music experiences around the globe," Bilal said.

Rather than wait for performance problems to emerge, Tuned Global reviewed how daily growth in new music releases and changing customer expectations would affect platform demands. It pointed to an industry environment in which more than 100,000 new tracks are released each day.

Scale gains

The modernised system increased music ingestion capacity from 2 million to 20 million tracks a day when onboarding new customers. Its extract, transform and load workflows can process up to 10 million tracks a day during large onboarding events.

That has shortened the time needed to onboard major catalogues. Music catalogues from groups such as Universal Music Group, Warner Music Group and Sony Music Entertainment, which previously could take up to two weeks to onboard, can now be processed in one to two days.

Tuned Global also reported database query performance that is three to five times faster, with support for more than 10,000 database read and write operations a second. At the application layer, the platform can scale from one million to 10 million API requests a minute while maintaining response times below 200 milliseconds.

The architecture uses Amazon Aurora PostgreSQL to scale reader and writer capacity independently, while Amazon Aurora Serverless adjusts capacity in line with demand. AWS Glue-based distributed processing also allows multiple catalogue ingestion jobs to run simultaneously rather than in fixed batches.

"Instead of having engineers adjust compute during ingestion peaks, we now rely on infrastructure that scales automatically," Bilal said.

The changes reduced infrastructure costs by 33 per cent and cut human intervention across ingestion and scaling workflows by about 25 per cent. In the AWS case study, Bilal said: "Moving to Amazon Aurora PostgreSQL gave us the scalability we needed to accelerate ingestion and customer onboarding while maintaining the performance standards our enterprise clients expect."

AI foundation

Bilal framed the project as part of a broader shift in how Tuned Global is preparing for AI-enabled music services. He said the company sees platform and data architecture as the starting point, rather than treating AI as a layer added later.

"There's a difference between legacy cloud and cloud-native. When we talk about legacy cloud and basically modernising that system, it's like having people living in a house and then doing renovations. You can only do so much based on your needs, but you can't really change the foundations of the house," Bilal said.

"Whereas with native AI-based solutions, you're redesigning everything from scratch. And that's how we started thinking. We thought about being cloud-native for AI, what does that look like? What do we need to change? What do we need to modernise? What are the partners out there that can help us with that?" Bilal said.

He said the updated architecture gives the company more flexibility to work with embedding data and to train proprietary audio models, including systems used for audio tagging and recommendations.

"That's an easy one. The platform comes first for us because you can add AI to a platform, but you cannot add a platform to an AI model, so to speak," Bilal said.

Bilal also said many businesses remain at an experimental stage in generative AI despite a wave of product announcements.

"I think in terms of the real end value, most enterprises are still stuck in a prototype phase, which means the demos are done, press releases are out, but there's no real transformation just yet," Bilal said.

Tuned Global is examining AI in terms of business outcomes and the data structures needed to support them. The company is already using AI-assisted tools in reporting, content curation, tagging and ingestion, and has expanded its AI-generated music detection work after acquiring UK-based audio AI company Figaro.ai.

Real-time demands

Bilal said the spread of agentic AI will make real-time, event-driven systems more important, especially in sectors where latency directly affects user experience.

"I think the gap is real and it's often underestimated. Most systems are not designed for real-time, event-driven feedback loops, for example. They're designed for batch processing, overnight runs to basically get an output and then optimise your systems based on that output," Bilal said.

In music services, he said, delays of even a fraction of a second can affect streaming, search and recommendations.

"In our domain specifically, we need real quick feedback loops. So, a 200 millisecond lag, for example, in streaming, search or recommendations, that is the difference between a seamless user experience and churn," Bilal said.

Bilal said the company's focus remains on data quality and organisational readiness as much as software tools. "So, cloud-native for us isn't just a technical choice. It's an organisational permission structure for AI," he said.