ML Software Engineer II – Embeddings & Multimodal AI

Google; Zürich; Presencial; CHF 150.000 - 190.000; Jornada completa

Descripción de la vacante

Google Zurich is seeking a software engineer to build scalable vector representations and two-tower models for shopping search. You will work on embeddings for queries, products, brands and categories, starting from a Gemini-based encoder, and contribute to end-to-end training and evaluation with large-scale data.

You'll optimize latency, participate in indexing product listings and multimodal embeddings, and collaborate across teams to push AI-powered experiences in a fast-paced environment.

Formación

• Bachelor’s degree or equivalent practical experience.
• 1 year of experience with software development in one or more programming languages (e.g., Python, C, C++, Java, JavaScript).
• 1 year of experience with data structures or algorithms.
• Experience integrating generative AI tools or LLM interfaces into workflows.

Responsabilidades

• Develop universal vector representations (embeddings) of products and related items, such as queries, brands, and categories.
• Build two-tower (dual-encoder) Transformer models that place queries and products in one vector space, utilizing a shared encoder that starts from a pretrained Gemini model.
• Train models on extensive datasets, utilizing user clicks and relevance labels distilled from a larger model.
• Execute training runs on shared Gemini infrastructure, encoding queries online with low latency while embedding and indexing product listings offline for fast nearest-neighbor vector search in the shopping search backend.
• Optimize multimodal Gemini embedding models that encode product thumbnails together with titles and descriptions to improve recall.

Conocimientos

Software development experience

Data structures & algorithms

Gen AI tools / LLM integration

Deep Learning / IR / Multimodal ML

Educación

Bachelor’s degree or equivalent practical experience

Descripción del empleo

Google Zurich is seeking a software engineer to build scalable vector representations and two-tower models for shopping search. You will work on embeddings for queries, products, brands and categories, starting from a Gemini-based encoder, and contribute to end-to-end training and evaluation with large-scale data.

You'll optimize latency, participate in indexing product listings and multimodal embeddings, and collaborate across teams to push AI-powered experiences in a fast-paced environment.