Platform - ML R&D - Principal Research/Data Scientist

What is The Role

The R&D ML team is responsible for training, testing and optimizing a set of models and pipelines used across the whole stack. As a model architecture lead for retrieval and reranking, you will lead the design and post-training strategies of our next generations of in-house models that will be integrated in our product. While our current generation of models are text only, we plan to add multi modality in the near future.

This is an outstanding opportunity to lead our effort and contribute to shipping the upcoming generation of Elastic’s models, which will then be served to our broad userbase.

What You Will Be Doing

• Strategic research & technical roadmap definition: identify promising new technologies for text-only and multi-modal models, look into post-training methodologies relevant for retrieval and re-ranking, and champion solutions that will fit best in Elastic.
• Tooling: improve our infrastructure and help scale to accommodate larger models: provide expert guidance on distributing data preparation, training and testing on multiple GPUs, help design, improve and write specs for our compute infrastructure
• Model variants exploration: investigate applying fine-tuning, quantization, distillation and other techniques to existing or upcoming models.
• Technical leadership: foster a culture of innovation and rigorous experimentation, coordinate knowledge sharing within the team; act as a senior technical authority and go-to expert for the modeling, training and testing challenges; mentor more junior members of the team helping them develop their skills and knowledge
• Communication: interact with various leads across the organization, communicate sophisticated concepts to high profile stakeholders, adjust and iterate on the roadmap according to internal and external inputs.

What You Bring

• Excellent comprehension of the industrial and academic state of the art on xLMs, more specifically in retrieval and reranking techniques
• Extensive professional experience in designing, training, and debugging various model architectures as well testing and implementing post-training strategies; proven track record of assessing, proposing and integrating frontier techniques to large scale products.
• 6+ years of professional software development experience in Python (pytorch).
• Experience with large-scale model training (between 0.1 and 7B parameters), distributed training, and model optimization techniques (e.g., quantization, pruning, efficient attention mechanisms).
• Excellent communication skills and gentle approach to leadership
• Strong attention to detail and highly organized