The RoleEvery customer's data warehouse is a mess in its own way. Tables nobody documented, three competing definitions of ARR, and a data team with an hour a week to give you. Our agents only work if that mess gets resolved into a clean representation of the business - and until it is, nothing built on top of it is trustworthy.
That's this job. You own the customer instance end to end: scoping data access, exploring the warehouse, mapping a chaotic schema onto real business concepts, standing up the ontology, building the signals on top of it, and proving the workflows moved something both sides agree on. You're customer-facing throughout and you report into engineering.
What You'll Do- Onboard new customers end to end. Scope data access, explore the warehouse with no documentation, and map a chaotic schema onto real business concepts.
- Stand up the ontology and own the gate before it. Interview the customer's data team and business owners, resolve their own inconsistent definitions, and validate the mapping before anything gets built on top of it.
- Build the signals on top of it. The predictions and workflows the customer actually sees.
- Define and measure success for each deployment. Agree the number with the customer up front, then report it back honestly whether or not it moved.
- Make the next deployment faster. Validation frameworks, eval harnesses, mapping accelerators - what you learn at one customer ships as tooling rather than tribal knowledge.
What You'll Bring- 3+ years in a technical, customer-facing role (forward deployed, solutions, or software engineering with heavy customer exposure); former founders strongly encouraged
- Deep data warehouse fluency. SQL, dbt, and real comfort in an undocumented five-year-old schema with nobody left to ask.
- You can sit with a customer's data or ops team, extract what a business concept actually means to them, resolve the cases where their own people disagree, and encode it.
- Applied data science literacy. Distributions, seasonality, validation design, calibration. You know when a result is noise.
- Painstaking through to the artifact. You'll chase a broken join for three hours because a number looked slightly off, and you'll care as much about whether the customer can actually interpret the output. Correct data in an unreadable form isn't delivered.
- You define success metrics rather than report on them. You can tell a customer their number didn't move and why.
- You hold a technical position under pushback from a customer or from us, and change your mind for evidence rather than authority.
- Production fluency with agent-driven development - Claude Code or equivalent as your default working mode
- High tolerance for ambiguity and unglamorous work. Based in NYC, excited to be in-office.
