<div class="show-more-less-html__markup show-more-less-html__markup--clamp-after-5 relative overflow-hidden"> On behalf of one of our clients in the fashion/retail sector, we are currently seeking a <strong>ML Engineer (m/f/d)</strong> for an upcoming project.<br/><br/><ul><li>Start Date: ASAP</li><li>Duration: until 31.12.2026 (with the option to extend)</li><li>Capacity: 100% (5 days per week)</li><li>Location: 100% Remote (Client Headquarter in Germany)<br/><br/></li></ul><strong> Projektbeschreibung <br/><br/></strong><strong>The problem you are solving<br/><br/></strong>Our client customer and product embedding models are in production and generating business impact already in personalized product retrieval in sorting. The infrastructure and architecture holding our client up — training pipelines, feature loading, model training time— needs to be rebuilt or adjusted for scale and speed. Our client also need to lay the groundwork for real-time inference.<br/><br/><strong>Specifically, You Will Work On One Or More Of<br/><br/></strong><ul><li>Reducing training data load times, ensuring train/serve consistency, enabling feature reuse across model versions of a two-tower neural Network </li><li>Horizontal scaling of TensorFlow training, including GPU utilisation optimisation </li><li>Inference &amp; Training pipeline: reliable, low-latency serving of embeddings and rankings at scale </li><li>Real-time personalisation groundwork: architecture for updating customer representations at inference time without full model retraining — requires deep understanding of how embeddings work and how they can be modified at serving time </li><li>Suggest and implement architectural changes in the model to improve quality <br/><br/></li></ul><strong>Skills<br/><br/></strong><ul><li>Proven experience building ML infrastructure in production — not prototype pipelines, but systems that serve real traffic reliably </li><li>Deep TensorFlow and Pytorch production experience (not just training scripts) </li><li>Hands-on experience with training and operating large neural networks at scale </li><li>Specific knowledge of embedding-based systems and low-latency inference is strongly preferred for the real-time workstream</li></ul> </div>