The world’s most popular audio streaming subscription service is looking for a Machine Learning Engineer to join the band in a consultant assignment. The client has transformed music listening forever when launched in 2008.

Period: ASAP to 2026-06-14 (full-time), with a possibility of extension.

About the role

Focus on these products:
QUAIL. Building and applying ML models for language evaluation, quality scoring, and multimodal translation workflows.

MAP (Multilingual AI Platform). It is a platform to translate content at scale using AI. MAP is designed to help efficiently localize and internationalize content for users in different languages and regions.

Who you are

• Expertise in NLP and LLMs (prompting, fine-tuning, evaluation). QUAIL depends on LLM-as-a-judge scoring frameworks.
• Experience with quality evaluation frameworks: (they have the rubrics designed, but it would be good if you have experience designing rubrics for language formatting; building evaluation APIs)
• Experience with data engineering for ML: collecting, cleaning, and structuring multilingual corpora for benchmarking.
• Comfort with MLOps: deploying models into production, monitoring drift, ensuring cost-efficiency.

Nice-to-Haves:

• Familiarity with MT evaluation: BLEU, COMET, MQM, or newer LLM-based scoring methods.
• Applied work in speech and multimodal ML: transcription (STT), speech isolation, diarization, TTS, dubbing.
• Ability to design scalable evaluation pipelines: run automated evaluations at segment/file/content level across multiple languages.
• Experience with human-in-the-loop evaluation workflows (sampling, quality control).
• Knowledge of cultural/linguistic nuances in translation (since QUAIL will measure “fit for purpose” quality).

We are Market Partner

Market Partner is proud to be an equal opportunity employer. You are welcome to our community regardless of who you are, no matter where you come from, or what you look like. We apply ongoing selection and may fill the position as soon as we find the right candidate.