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.
AI-First by Default
At Deepgram, we expect an AI-first mindset—AI use and comfort aren’t optional, they’re core to how we operate, innovate, and measure performance.
Every team member who works at Deepgram is expected to actively use and experiment with advanced AI tools, and even build your own into your everyday work. We measure how effectively AI is applied to deliver results, and consistent, creative use of the latest AI capabilities is key to success here. Candidates should be comfortable adopting new models and modes quickly, integrating AI into their workflows, and continuously pushing the boundaries of what these technologies can do.
Additionally, we move at the pace of AI. Change is rapid, and you can expect your day-to-day work to evolve just as quickly. If you’re energized by experimenting, adapting, and learning constantly, you’ll thrive here. Expect your day-to-day to evolve as fast as the technology does.
About the role
Deepgram is building its decision layer: one governed set of numbers that people and AI agents can act on without re-deriving them. This role builds it.
You are not building pipelines. Ingestion, orchestration and latency belong to the data platform team, and they're good at it. You take what lands in the warehouse and turn it into the layer above the conformed gold tables, the semantic models, and the agent-facing tools that let someone (or something) ask a question and get an answer that holds up.
This is a senior technical IC role on a new team. You'll be the first or second engineer in.
What you'll work on
Model the gold layer. Conformed entities for customer, account, project, product and contract. Our identity model spans Salesforce, the console, Stripe and the billing system, with no authoritative key between them resolving that is week one.
Own the semantic layer. Author the models that define our metrics once, so a dashboard, a notebook and an agent all return the same number. Version them. Retire the superseded ones.
Build the agent-facing surface. The tools, permissions and MCP surfaces that let analyst agents navigate our data without inventing a join that looks plausible and is wrong.
Make correctness checkable. Assertions on the things that break silently grain, uniqueness, cross-system reconciliation routed to a named owner. Evaluation of agent output against known-correct answers, tracking accuracy, refusal, and confidently-wrong rates separately.
Absorb the questions. Sales, product and finance bring real questions. You answer them, and you leave behind a model that makes the next version of that question cheaper.
What we're looking for
AI-native practice, demonstrated. You use AI daily and intricately not to autocomplete, but as a way of working that has changed what you attempt. We will ask what the most interesting thing you've built with AI in the last month is, and we'll go three questions deep on the answer.
Production-grade SQL and Python. Code that ships on a schedule, has broken, and was found by your own instrumentation rather than by a stakeholder.
Semantic or metrics layer ownership. dbt, Cube, LookML or equivalent owned, not used. You've changed a definition people already depended on, and can say who you told and what the number moved by.
Modelling judgement over tool knowledge. You reason about grain and keys before you write the query. You can tell a data defect from an undecided business definition and treat them differently.
Comfort in work-in-progress infrastructure. Nothing here is finished. There is no ticket queue to work from and no process to inherit you'll be building both as you go.
Clear writing. Much of the output is a definition or a contract someone else has to trust without re-deriving it.
Bonus Points
Usage-based or consumption business models, where committed, consumed, invoiced and recognised revenue are genuinely different numbers.
Real-time billing and consumption systems.
Lakehouse architectures - Iceberg, Athena, Trino or similar.
Contributions to open-source data or AI projects.
Notice: We're aware of individuals impersonating Deepgram recruiters. All legitimate Deepgram recruiting communication comes from an @deepgram.com email address. If you've received a message claiming to be Deepgram, please forward it to careers@deepgram.com.