Fabian Moerchen

Leading science at Amazon Music since 2017 helping grow the customer base from 16M to 80M. Setting the science strategy for a large team of scientists working on recommendations, search, conversations, content understanding, fraud detection, and retention modelling. Establishing collaborations across the company with Prime Video, Audible, Alexa, and AWS. Coaching scientists across teams on project execution and career growth. Most recently focused on customizing Large Language Models (LLMs) for the music domain and specific tasks such as recommendations, intent understanding, and conversations to offer high quality inference with low latency. At Amazon Music we are using artificial intelligence, machine learning, and large datasets comprised of audio data, meta-data, and customer behavioral data to improve your listening experience. We iterate rapidly using state-of-the-art tools and algorithms. We use a wide range of methods suited to each problem including LLM (CPT, SFT, GRPO, RAG, tools, agents), deep learning (FFNN, sequential models, auto-encoders), and traditional methods (bandits, causal models, classification, regression, hypothesis testing, clustering, pattern mining, signal processing). We work on long term high risk and high reward projects to power innovative future use cases and short term projects to improve the production systems that provide a voice-enabled and personalized listening experience for Amazon Music customers. In more than 20 years of working in the field of artificial intelligence and machine learning I have published more than 50 research papers and have given tutorials on temporal data mining at top conferences. My first paper was on training neural networks in parallel! In recent years at Amazon the publication volume is lower, as much of the research is only published internally or submitted as patents (https://www.amazon.science/a