Senior Analytics Engineer, Decision Intelligence

Deepgram runs speech-to-text and text-to-speech APIs for more than 200,000 developers, and this role builds the layer that tells the company what is actually happening in its own business.

The posting draws an unusually clear line around the job: you are not building pipelines. Ingestion, orchestration and latency belong to the data platform team, and this role takes what lands in the warehouse and builds everything above it. Concretely that is the conformed gold layer, with entities for customer, account, project, product and contract, and the identity model spanning Salesforce, the console, Stripe and the billing system with no authoritative key between them, which the posting says is week one. Then the semantic layer, so a dashboard, a notebook and an agent all return the same number, versioned, with superseded definitions retired. Then the agent-facing surface: the tools, permissions and MCP surfaces that stop an analyst agent inventing a join that looks plausible and is wrong. Correctness is checkable by design, including evaluation of agent output that tracks accuracy, refusal and confidently-wrong rates as three separate numbers.

It is the first or second engineer on a new team. Asks are production-grade SQL and Python, and ownership rather than use of a semantic or metrics layer, dbt, Cube or LookML. Read the operating-rhythm section before applying: Deepgram says AI use is not optional, that it measures how effectively you apply it, and that this is the wrong role if you want something prescriptive. Remote, US.