Build the system that reads a brand's evidence (decks, survey exports, sales data, retailer emails) and turns it into a product decision where every line is labelled and sourced.
What you'll do
- Evidence intake: pull clean, sourced facts out of PDFs, spreadsheets and email threads.
- Labelling: separate evidence, inference and assumption, and keep the source on every line.
- Evaluation: score calls against reviewer edits and real launch outcomes, on every change.
- Product: ship the decision record, the reviewer workspace and the quick read.
Who you are
- You've shipped LLM features that people pay for, and you know where they fail.
- You're strong in Python and TypeScript, and comfortable owning a feature end to end.
- You care more about being right than sounding confident.
- Retail, CPG or market research data is a plus, not a requirement.