The posting
COMPANY OVERVIEW
Deepgram is bringing the world’s most natural interface, voice, to the technology we use every day. Our speech models and real-time infrastructure power AI systems that keep up with the way people actually speak. More than 5,000 organizations run conversational voice AI on Deepgram in production, including Twilio, Cloudflare, Sierra, Decagon, Vapi, Cresta, Granola, and Jack in the Box.
The Deepgram Voice Agent Platform brings together speech-to-text, text-to-speech and speech-to-speech, with voice-native foundation models available through cloud APIs or self-hosted and on-premises deployments. Built for accuracy, low latency and cost efficiency, Deepgram makes high-quality voice experiences practical to build and run at scale. More than 650,000 developers have used Deepgram to process more than 200,000 years of audio. Every day, more of the world’s conversations with AI run on our models.
OPERATING PRINCIPLES
Deepgram is a research company that productizes as fast as possible. Four Operating Principles shape how we think, decide, and work together:
1. Compress before you communicate. You think it through first, then lead with the bottom line.
2. Resist entropy. You question what already exists before adding more process or tools.
3. Solve the category. You fix recurring problems at the source, not one case at a time.
4. Work at the frontier. You go after problems no one has solved yet.
We look for people who already work this way, and these principles shape how we grow and evaluate our team.
AI-FIRST BY DEFAULT
We build AI that millions of people talk to, and we run the company the same way. AI is how we centralize work here, and is core to how we operate.
That means more than using the tools. We expect you to:
- Build, not just use. Design your own workflows and agents, and use AI to remove whole categories of work instead of doing the same work faster.
- Own the output. AI can speed up the work. It doesn’t make anyone less accountable for it. You understand, verify, and stand behind everything you share, including what an agent produces for you.
- Spend the time you get back at the frontier. The capacity AI creates goes toward problems no one has solved yet.
- Keep pace. Models and tools change every month, and so will your work. You adopt what’s new quickly and drop what no longer earns its place.
We judge AI by what it helps you deliver and whether you stand behind the result, not by how often you use it. In interviews, expect to walk us through a workflow you built and where you drew the line between AI and your own judgment.
Team Overview
Deepgram’s Frontier Research Staff is a team of innovators building the foundational models behind our perception (speech-to-text) and generation (text-to-speech) products. Our bet: today’s approach of scaling models and data together won’t deliver voice AI that works for everyone. Real-world audio is scarce, enormously diverse, and expensive to train on at world scale, so we’re inventing new paradigms instead. You’d join a group that defines its own research agenda, ships what it learns into production, and moves at the pace of AI.
Key Goals
- Define and own new LLM research directions that advance Deepgram’s voice AI and LLM models https://developers.deepgram.com/docs/llm, choosing high-leverage bets together with the rest of Research Staff
- Turn the fast-moving LLM literature into leverage by surveying, distilling, and deciding which methods are worth building on
- Run fast, decisive experimental programs, finding the experiment that settles a question in days rather than months
- Take successful proofs-of-concept to production scale, driving distributed training runs through to models our developers and customers use
- Build durable research systems, such as data curation, training, and evaluation pipelines, that remove recurring manual work for the whole team
- Bring LLM advances into our voice products, partnering with Engineering and Product to turn research into better perception and generation
- Communicate results in a way that moves decisions, so research, engineering, and product leaders understand what was learned and why it matters
Minimum Skills, Knowledge & Capabilities
- Hands-on experience training LLMs end-to-end, including data curation, distributed large-scale training, transformer architecture optimization, and reinforcement learning (RL) post-training
- Deep working knowledge of transformer architectures (autoregressive, sequence-to-sequence, causal LMs): why they behave the way they do and how to change them
- Strong Python and PyTorch, with research code that holds up when it scales to real training infrastructure
- Distributed computing and large-scale data processing, including multi-node training and inference
- Designs experiments that cleanly validate or kill an idea, then turns the results into model improvements
- Works AI-native: designs Human+AI workflows for research (literature review, experiment generation, analysis) and automates their own pipeline instead of re-running it by hand
- Explains complex technical work clearly to researchers, engineers, and product partners
- PhD or Master’s degree in Computer Science, Machine Learning, Electrical Engineering, Applied Mathematics, or a related field, or equivalent industry research experience
Preferred Qualifications (nice-to-have, not required)
- Typically ~5 - 8 years in applied deep learning research, or equivalent demonstrated experience
- Published research in deep learning, particularly on LLMs or deep neural networks
- Familiarity with RLHF labeling and training pipelines
- Experience with distributed inference or efficient LLM serving
- Exposure to speech, audio, or multimodal models
We know the confidence gap and imposter syndrome can get in the way of meeting great candidates — so please don’t hesitate to apply. We’d love to hear from you.
Notice: We're aware of individuals impersonating Deepgram recruiters. All legitimate Deepgram recruiting communication comes from an @deepgram.com http://deepgram.com email address. If you've received a message claiming to be Deepgram, please forward it to [email protected].



