Applied AI Engineer

Plotwell; San Francisco (CA); On-site; USD 180,000 - 240,000; Full time

Job summary

Plotwell is seeking a software professional who will own the prompts and the JSON contract for chart selection and key takeaways. You will build evaluation sets with real dashboards and ensure the model remains agnostic while measuring cost-effectiveness across iterations.

You will collaborate with engineering to implement grounding checks that drop ungrounded numbers and demonstrate measurable improvements through repeatable experiments.

Qualifications

• Four or more years shipping software, with real language-model work behind you.
• You evaluate rather than vibe-check, and you can prove a change helped.
• You are comfortable owning a fuzzy problem with a measurable output.
• You want the model to be the smallest, best-scaffolded part of the system.

Responsibilities

• Own the prompts and the JSON contract for chart selection and takeaways.
• Build the evaluation set: real spreadsheets with known good dashboards, run on every change.
• Keep Plotwell model-agnostic, and measure when a cheaper model is good enough.
• Work with engineering on the grounding check that drops ungrounded numbers.

Skills

Shipping software

Language-model work

Evaluation mindset

Model-agnostic approach

Job description

When Plotwell looks at a file and decides the story is revenue by month with churn under it, that judgement comes from a model and a lot of careful scaffolding. You own that.

What you would own

• Own the prompts and the JSON contract for chart selection and takeaways.
• Build the evaluation set: real spreadsheets with known good dashboards, run on every change.
• Keep Plotwell model-agnostic, and measure when a cheaper model is good enough.
• Work with engineering on the grounding check that drops ungrounded numbers.

What we are looking for

• Four or more years shipping software, with real language-model work behind you.
• You evaluate rather than vibe-check, and you can prove a change helped.
• You are comfortable owning a fuzzy problem with a measurable output.
• You want the model to be the smallest, best-scaffolded part of the system.

How hiring runs

A thirty minute call with a founder, then a paid work session on a real problem from our backlog, three to four hours, then conversations with two people you would work alongside. Four steps, usually inside two weeks, and we tell you where you stand at every one.