What the work actually looks like
AfterQuery contracts domain experts to produce training and evaluation data for frontier AI labs. On a finance project, that means writing tasks a model should be able to do — a three-statement build from a filing, an LBO with a debt schedule, a comps table with the adjustments spelled out, a short investment memo — and then writing the reasoning trace that gets from the raw inputs to the answer. The second half of the job is evaluation: you read a model's Excel output or memo and judge whether the numbers tie, whether the structure resembles what a deal team would actually circulate, and whether the logic holds or merely sounds plausible.
Most deliverables arrive as an Excel workbook, a PowerPoint page, or a written rationale, with a rubric or annotation form attached. Expect to defend your grading — project leads and other reviewers will ask why you marked a valuation wrong, and "it felt off" will not survive that conversation. The most useful contributors are the ones who can point to the specific line item, circularity, or unsupported assumption that breaks the model.
What the screen is looking for
The qualification that carries the most weight is a completed internship at a bulge bracket or elite boutique bank, a top-tier private equity fund, or a hedge fund. The screen probes that hard: what you built, in what software, for which deal or coverage group, and what your associate sent back for revision. Second, it tests whether you can decompose finance into steps — not summarise a DCF, but narrate the order of operations and say where a junior analyst usually goes wrong. Vagueness reads as coursework rather than desk experience.
Logistics
- Fully remote and asynchronous; no fixed hours, but batches carry deadlines.
- Roughly 10–15 hours per week when you are staffed on an active project; volume is uneven between projects.
- Contract work, project-based. Junior and senior undergraduates studying finance or economics are explicitly welcomed alongside early-career analysts.
- Excel fluency is assumed and may be checked with a build task rather than a conversation.