QA Engineering Intern (AI Product)
Trendline is a real-time terminal for prediction markets — one screen for live prices, order books, and charts across Kalshi, Polymarket, and other venues. Optimus is the analyst built into it. It answers plain-English questions from two sources: live market data, and years of historical data across every domain the markets cover — sports results, election outcomes, economic releases, past market prices.
Each source fails differently, and both look right when they fail. On the live side: a stale price quoted as current, or two similar markets confused. On the historical side: a record pulled from the wrong subset — home games when you asked about road games, a candidate's primary margin when you asked about the general. And between the two sits the hardest class: questions where the wording carries the meaning — markets that sound identical but resolve on different criteria, and an answer that's accurate for the question it heard rather than the one you asked. We're hiring someone whose whole job is finding these before users do.
3 months · $3,000 stipend · remote · flexible hours, judged on output.
What you'll do
- Question Optimus the way real users will, in volume, and log where it's wrong, vague, or slow.
- Check its historical claims against the actual record — when it cites a win-loss split, an election result, or a past price, verify the subset, the period, and the arithmetic.
- Write reproductions an engineer can act on the same day: the question, the answer it gave, the correct answer, and the evidence.
- Convert every confirmed failure into a regression test, so nothing breaks twice.
- Maintain the demo question set — know exactly which questions hold up in front of an audience.
- Classify failures by cause (stale data, wrong market, wrong subset, misread question, bad formatting) so engineering fixes the most expensive class first.
What we're looking for
- You notice when a number looks off, and you check it.
- You know at least one of these worlds — sports, politics, markets — well enough to smell a wrong number without looking it up.
- You read questions closely; a lot of this job is catching answers that are right for the wrong question.
- You write clearly — precisely enough that someone else can fix what you found.
- Patience with bugs that don't reproduce on the first try; against live data, some failures only appear at the wrong moment.
- No CS degree required — the domain and tooling are teachable.