Machine Learning Research Engineer - Search and Retrieval

Overview

Perplexity is recruiting a seasoned Machine Learning Research Engineer to develop cutting-edge search technologies focused on retrieval and ranking capabilities. The role involves advancing search quality through innovative modeling, data utilization, and tool development.

Responsibilities

• Drive continuous improvements in search quality via models, data, tools, and other resources.
• Design and develop core components of the search platform and model architecture.
• Create, train, and fine-tune large-scale deep learning models with frameworks such as PyTorch, employing distributed training methods like PyTorch Distributed, DeepSpeed, and Fully Sharded Data Parallel (FSDP), emphasizing retrieval and ranking.
• Engage in advanced research on representation learning, including contrastive learning and multilingual/multimodal modeling relevant to search and retrieval.
• Deploy scalable and efficient models ranging from boosting algorithms to large language models (LLMs).
• Develop and enhance Retrieval-Augmented Generation (RAG) pipelines for information grounding and answer generation.
• Work closely with teams across Data, AI, Infrastructure, and Product to ensure rapid and high-quality deployment and delivery.

Qualifications

• In-depth knowledge of search and retrieval systems and their quality evaluation metrics.
• Demonstrable experience with large-scale search or recommendation systems.
• Advanced proficiency with PyTorch, including distributed training and performance tuning of large models.
• Expertise in representation learning techniques such as contrastive learning and embedding alignment for multilingual and multimodal contexts.
• Strong publication history in prestigious AI/ML conferences or workshops (NeurIPS, ICML, ICLR, ACL, CVPR, SIGIR).
• Highly self-motivated with a strong sense of accountability and drive for execution.
• A minimum of three years, preferably over five years, of experience in search, recommender systems, or closely related research domains.