Senior Machine Learning Engineer - Recommendations (Core)

We are looking for a Senior Machine Learning Engineer to join ML Core, the team that builds the shared ML models behind SoundCloud’s recommendations, search and content understanding. New sounds start on SoundCloud. Our job is to connect artists with the communities that will care about their music, and help listeners discover the gems and emerging scenes they’d never find on their own. Recommendations are our largest area, but you’ll work across problems that share the same foundations: representing users, tracks and creators, and ranking them well. We’re also in the middle of modernizing our stack, moving from classical approaches to deep learning on GPU, and you’ll play a big part in that.

• →Design, train and ship models for recommendations (candidate generation, user/item modeling, ranking) and related problems in search and content understanding.
• →Lead the move of key models from classical approaches (e.g., word2vec-style embeddings) to PyTorch-based models trained and served on GPU.
• →Adapt, fine-tune and evaluate open-source embedding and ranking models for our catalog and use cases.
• →Own evaluation end to end: offline metrics, A/B experiments, and making sure offline gains show up online.
• →Work with ML, data and backend engineers to get models into production and keep them healthy, and help set modeling standards across the team.

• 3+ years building ML systems in production, with end-to-end ownership of at least one ranking, retrieval or recommendation model.
• Strong deep learning skills in PyTorch (or similar frameworks), and a good grasp of recommendation and retrieval fundamentals (embeddings, two-tower models, learning to rank, multi-objective ranking) and their trade-offs.
• Solid experience with evaluation and experimentation for ranking systems, including biases in implicit feedback such as clicks and plays.
• Strong Python; comfortable with large-scale data processing (e.g., Spark/BigQuery) and with getting models to production on a cloud platform (GCP preferred).
• M.Sc. in a quantitative field or equivalent experience.

• Fine-tuning or serving transformer-based embedding/ranking models.
• Applying LLMs or foundation models to recommendations (content understanding, semantic IDs, generative retrieval, feature generation).
• Audio or music ML, or multimodal representations.
• Contextual bandits, RL or causal methods for long-term value.
• Publications at RecSys, KDD, SIGIR, WSDM, NeurIPS or similar.
• Experience with ANN/vector search and GPU training or serving at scale.

We encourage you to apply even if you don't meet every qualification. Research suggests that some applicants may doubt their candidacy, so please submit an application if you are interested.