Remote Research Scientist - Voice AI Foundations

Summary

Join Deepgram, a foundational AI company, as a Research Scientist and contribute to transforming human-machine interaction using natural language. You will design experimental programs, drive large-scale training jobs, optimize model architectures, and collaborate with product & engineering teams.

Requirements

• PhD in Physics, Electrical Engineering, Computer Science or another related field
• Prior experience in designing and conducting experimental programs aimed at understanding complex phenomena, with the ability to rapidly iterate and change course as needed
• Proven experience building models from a blank page and owning the entire deep learning stack including data curation, characterization and cleaning, architecture design and model building, distributed large-scale training, and model optimization for inference
• Strong communication skills and the ability to translate complex concepts in simple terms, depending on the target audience
• Strong software engineering skills with particular emphasis on developing clean, modular code in Python and working with Pytorch

Responsibilities

• Design and carry out experimental programs to build new speech and language AI foundation models across modalities and tasks, that solve critical problems for our customers
• Drive large-scale training jobs successfully on massive distributed computing infrastructure
• Optimize model architectures to make them as fast and memory-efficient as possible; deploy new models into production for use at massive scale
• Document and present results and complex technical concepts clearly for internal and external audiences

Preferred Qualifications

• Prior industry experience in building deep learning models to solve complex problems, with a solid understanding toward the applications and implications of different neural network types, architectures, and loss mechanisms
• Deep understanding and experience working with state-of-the-art network architectures including transformers
• Understanding of different parallelism paradigms for efficient distributed training