Meet Gunay Abdullayeva, a Machine Learning Engineer at Epidemic Sound, whose work helps power the smart discovery features behind our platform. From idea to product release, she’s involved in every step of the ML lifecycle—turning ideas into impactful tools for our users. In this interview, Gunay shares her journey, how ML and other teams collaborate, and what makes working with machine learning at Epidemic truly unique.

My name is Gunay Abdullayeva and I have been a Machine Learning Engineer at Epidemic Sound since August 2021. I come from Azerbaijan and I got my bachelor in Computer Science and later on my Masters in Machine Learning in Estonia. 

I’ve mainly worked in ML, however, before ES, I worked as a Data Scientist within the banking field. I switched over to Back-End Engineering to explore more on software building. Soon realised that I missed ML and I enjoy building software and I wanted to do both! So I decided to proceed on ML engineering where you get both.

Outside of my profession you can find me enjoying nature! I absolutely love biking, hiking and camping. 

When I was applying to Epidemic Sound, I was also in the recruitment process with other companies. But from my very first call with Max, the recruiter, I could feel the difference. Throughout the process, I had the chance to virtually meet more team members, dive into the product, and get a real sense of the role. It wasn’t just exciting—it felt like fun!

I was hoping to move to Germany as I had friends over there, I never planned to move to Sweden but Epidemic was just too nice to say no to. 

I work within our Discovery team which primarily focuses on discovery features for our users. Some recent features that I have worked on are our AutoSpot-SFX feature which recommends sound effects based on your video content and our Semantic search model which enhances our search engine to understand more prompt-like searches.  

We build and deploy all our services in Google Cloud Platform (GCP), leveraging a range of its machine learning services especially Vertex AI for model training. Our data is primarily stored in BigQuery and Cloud Storage. We also work extensively with Kubernetes, particularly for building and deploying APIs for our users. In terms of programming languages, Python is our primary choice, but within my team, we also use Go for API development—which I’ve heard is quite fun!

We collaborate closely with our UI/UX research team, who regularly engage with users to gather feedback on different features—what they like, what they don’t, and what they wish they had. Over time, they consolidate this feedback, identifying common themes and prioritizing what matters most to our users. This helps us make informed decisions about our product direction. Ultimately, our work is driven by both user insights and industry trends.

Backend engineers and ML engineers collaborate closely, though the dynamics can vary depending on the project.

When a new feature idea emerges whether or not it's initially considered an ML-driven feature both backend and ML engineers work together to explore the best way to implement it. Sometimes, what starts as a standard feature evolves into an ML-powered solution as they dive deeper into the problem.

The process typically begins with the ML engineer conducting a study phase, researching available tools, setting up datasets, training models and lastly serving the model in an API. During this phase, the backend engineer is less involved. However, once the model is trained and served, collaboration intensifies. The backend engineer takes the lead in integrating the deployed model into the system, making architectural decisions, and ensuring the model serves users effectively.

While the level of involvement differs at each stage, the key to a smooth collaboration is continuous alignment—figuring out together how to bring the best possible solution to life.

In ML projects, my involvement spans the entire lifecycle from problem definition to model deployment and serving.

Everything starts with identifying a problem and figuring out a solution. Once the direction is clear, the hands-on work begins with shaping the dataset. This step is crucial, as the dataset must be structured correctly to train the future model effectively.

From there, I work on building the initial model, evaluating its performance using historical data and relevant metrics. We also evaluate through manual testing, running API tests, analyzing inputs and outputs, and ensuring the results make sense.

Since model development is rarely a one-and-done process, this evaluation often leads back to refining the dataset, identifying missing information, collecting more data, and improving features. This cycle continues until the model is strong enough to be deployed and tested with real users.

Ultimately, I’m involved in every phase, iterating until we reach a point where the feature is ready to be shipped and evaluated in a real-world setting.

It’s the level of involvement and freedom ML engineers have throughout the entire process that stands out here. 

ML engineers don’t just execute ideas—they can also originate them. If you notice an opportunity or a problem others might not see, you have the space to propose and explore solutions. 

We are also engaged in every stage of development, from shaping the dataset to collaborating with data scientists, backend engineers, and designers. This cross-functional approach ensures that ML features are not only technically sound but also well-integrated into the product experience.

There’s a strong culture of experimentation and learning. While there are defined goals, ML engineers have the flexibility to test ideas, participate in hack weeks, and even explore other domains. This open and collaborative environment makes ML at Epidemic both impactful and dynamic.

My first successful A/B test! I was working on the new model to improve Soundmatch results. We launched the A/B test and we got 2.5% uplift on the search success which was great to see!