SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Deepgram is building its decision intelligence layer—a governed set of metrics and models that people and AI agents can reliably act on without re-deriving them. This is a senior individual contributor role on a new team, and you'll be the first or second engineer building it.
You will not be building data pipelines; the data platform team owns ingestion, orchestration, and latency. Instead, you take what lands in the warehouse and build the semantic layer above conformed gold tables—the models and agent-facing tools that let someone (or something) ask a question and get an answer that holds up.
Key responsibilities:
- Model the gold layer: define conformed entities for customer, account, project, product, and contract. Resolve identity across Salesforce, the console, Stripe, and the billing system where no authoritative key currently exists.
- Own the semantic layer: author metrics definitions once so dashboards, notebooks, and agents all return the same number. Version and retire superseded models.
- Build agent-facing surfaces: create tools, permissions, and MCP interfaces that let analyst agents navigate data safely without inventing plausible-but-wrong joins.
- Make correctness verifiable: implement assertions on grain, uniqueness, and cross-system reconciliation routed to named owners. Evaluate agent output against known-correct answers, tracking accuracy, refusal, and confidently-wrong rates.
- Absorb real questions from sales, product, and finance. Answer them and leave behind a model that makes the next version cheaper.
The role operates in an AI-first environment. Deepgram expects all team members to actively use and experiment with advanced AI tools daily, integrating them into workflows and continuously pushing boundaries. Change is rapid, and you should expect your day-to-day work to evolve quickly. This is not a prescriptive, traditional 9-to-5 role.
Requirements:
- Demonstrated AI-native practice: you use AI daily and intricately as a way of working, not just for autocomplete. Be prepared to discuss the most interesting thing you've built with AI in the last month.
- Production-grade SQL and Python: code that ships on schedule, has broken in production, and was caught by your own instrumentation rather than a stakeholder.
- Semantic or metrics layer ownership: hands-on experience with dbt, Cube, LookML, or equivalent. You've owned and changed definitions that people depended on; you can articulate who you told and how the numbers moved.
- Strong modelling judgment: you reason about grain and keys before writing queries. You distinguish data defects from undecided business definitions and treat them differently.
- Comfort with work-in-progress infrastructure: nothing here is finished. There is no ticket queue or inherited process; you'll build both as you go.
- Clear writing: much of your output is definitions or contracts others must trust without re-deriving.
Bonus qualifications:
- Experience with usage-based or consumption business models where committed, consumed, invoiced, and recognized revenue are genuinely different numbers.
- Real-time billing and consumption systems.
- Lakehouse architectures (Iceberg, Athena, Trino, or similar).
- Open-source contributions to data or AI projects.
