Research Staff, LLMs

Deepgram Inc.

Job Overview

Location

USA | Remote

Job Type

Full-time

Full Job Description

📋 Description

• Collaborate with other Research Staff members to brainstorm and define new large language model (LLM) research initiatives focused on advancing voice AI capabilities.
• Conduct broad literature surveys, evaluate, classify, and distill current methods in deep learning and LLMs to inform research direction.
• Design and execute experimental programs for LLMs, including testing novel architectures, training methodologies, and data curation strategies.
• Drive transformer-based LLM training jobs on distributed compute infrastructure, ensuring efficient resource utilization and successful model convergence.
• Deploy trained LLM models into production environments, ensuring scalability, low latency, and compatibility with Deepgram’s voice-native APIs.
• Document research findings and present complex technical concepts clearly to both technical and non-technical audiences.
• Stay current with the latest advances in deep learning, particularly in transformer architectures, reinforcement learning, and LLM optimization techniques.
• Apply AI tools and automation to accelerate research cycles, consistently seeking ways to amplify personal and team impact through AI-driven workflows.
• Identify critical experiments that can validate or refute hypotheses within days, not months, to maintain rapid iteration cycles.
• Scale successful proofs-of-concept by 100x, transforming experimental results into production-ready components for voice AI systems.
• Work with high-dimensional, real-world audio data to address core challenges in scarcity, diversity, and computational cost in voice AI training.
• Optimize transformer architectures for efficiency, including auto-regressive and sequence-to-sequence models, to improve performance and reduce inference costs.
• Leverage reinforcement learning techniques, including RLHF pipelines, to align model outputs with human preferences in voice interaction.
• Contribute to the development of new data curation frameworks tailored to the unique demands of audio-based LLM training.
• Engage in continuous learning through participation in internal AI enablement workshops, external conferences, and research talks.
• Maintain a rigorous, analytical approach to model evaluation, using detailed analysis to drive iterative improvements in LLM performance.
• Build new systems from the ground up, prioritizing elegant, scalable solutions to fundamental problems in voice AI.

🎯 Requirements

• 3+ years of experience in applied deep learning research with understanding of neural network architectures and loss mechanisms
• Proven experience working with large language models (LLMs), including data curation, distributed large-scale training, transformer optimization, and reinforcement learning
• Strong coding proficiency in Python and experience with PyTorch
• Experience with various transformer architectures (e.g., auto-regressive, sequence-to-sequence)
• Experience with distributed computing and large-scale data processing
• Prior experience conducting experimental programs and using results to optimize models

🏖️ Benefits

• Medical, dental, and vision benefits
• Annual wellness stipend
• Mental health support
• Unlimited PTO
• Generous paid parental leave
• Flexible schedule
• 12 paid US company holidays
• Quarterly personal productivity stipend
• One-time home office upgrade stipend
• 401(k) plan with company match
• Learning/education stipend
• Participation in talks and conferences
• AI enablement workshops and sessions

Skills & Technologies

Python

PyTorch

Senior

Remote

Ready to Apply?

Apply Externally

You will be redirected to an external site to apply.

Deepgram Inc.

Visit Website

About Deepgram Inc.

Deepgram builds end-to-end speech AI infrastructure that converts live or recorded audio into text and insights. The company trains large-scale neural networks on GPU clusters to deliver low-latency transcription, keyword detection, and speaker diarization through a single API. Developers use the platform for call centers, meetings, podcasts, and voice bots, paying per minute or hosting the engine on-premise. Founded in 2015 and headquartered in San Francisco, Deepgram serves enterprises seeking accurate, private, and customizable speech recognition without vendor lock-in.

View Company Profile

Get more remote jobs like this

Subscribe to the weekly newsletter for similar remote roles and curated hiring updates.

Weekly remote jobs and featured talent.

No spam. Only curated remote roles and product updates. You can unsubscribe anytime.