On a music streaming platform, the search engine plays a central role in helping users access and discover content. While many users search with a clear intent—such as playing a specific song or artist—others use search in a more exploratory way, looking for new or personalized content through open-ended queries like “chill indie vibes” or “road trip songs from the 2000s.”

Traditional search engines, which rely on lexical or semantic matching, often struggle with this kind of exploratory intent. Recommender systems can help but are not always designed to handle complex, free-form queries. This is where Large Language Models (LLMs) and agentic technologies come in: they can interpret nuanced user intent and decide how to best retrieve relevant content by dynamically combining search, recommendation, and other retrieval strategies.

The internship has a start date in October and the goal is to explore how these agentic approaches—powered by LLMs—can be applied to improve exploratory search on a music platform. You will help design and prototype an intelligent system that understands user queries and routes them to the most appropriate back-end services (e.g., search index, recommender, curated playlists).

What you’ll work on

• Investigate and evaluate recent research on agent-based and LLM-powered search systems

• Analyze the specific needs of a music-focused search engine, which differs from general-purpose search

• Design and implement a prototype that interprets complex user queries and delivers personalized music results

• Work with real-world usage data from the Deezer platform

• Optionally experiment with fine-tuning or prompt-engineering LLMs for better performance

• Contribute to system evaluation and discuss the path toward scaling and production deployment

Who we’re looking for

• Master’s student in Computer Science, Machine Learning, or a related field

• Solid understanding of natural language processing and machine learning and good coding skills in Python

• Familiarity with LLMs (e.g., Transformers, OpenAI models) and retrieval techniques (e.g., Elasticsearch, FAISS) is a strong plus

• Knowledge of cloud environments (e.g., AWS, GCP) is a plus

• Interest in user-centric search and recommendation systems

• Curiosity, autonomy, and a hands-on approach to prototyping and experimentation

If you don’t meet 100% of the qualifications outlined above, tell us why you’d still be a great fit for this role in your application!