About
About Aethra & our research
What we're building, and what our research has and hasn't shown. Last updated September 22, 2026.
What Aethra is
Aethra Copilot is a set of AI tools for job seekers: a resume analyzer with an ATS score, a resume optimizer, cover letter and LinkedIn tools, a job tracker, and a community feed. It’s currently in beta.
The question we're researching
AI career advice sounds equally confident whether it’s right or wrong. A suggestion like “add these keywords” or “lead with this project” reads the same either way — and usually nobody goes back to check whether it actually helped.
So the question we’re working on is simple to state: does career advice that’s checked against what really happened to people — did they get the interview? — before it’s trusted, do better than advice that isn’t checked?
What we've shown so far
To answer that honestly you need a way to measure how a different set of advice would have performed, using only records of the advice that was actually given. We’ve built that measuring method (a Direct Method estimator), and validated it in two steps before trusting it on anything real.
First, a sanity check: we simulated a world where we already knew the true answer — a recommendation that truly performs 0.050 better, logged by a policy deliberately biased toward picking it in favorable situations. The estimator recovered 0.046, close to the true 0.050 and clearly better than a naive before/after comparison (which was off by more than 3x). Repeated across 40 simulated worlds, its confidence interval contained the true answer 37 times; across 40 worlds with no real effect, it wrongly claimed one 3 times — in line with what a correctly calibrated method should do.
Second, a real-world test using the Open Bandit Dataset, a public dataset from a real online fashion store that ran two recommendation approaches side by side. We gave our method the records of only one approach — more than 12 million of them — and asked it to predict how often shoppers clicked on the other one. Across three tests, its predictions landed within 0.1%, 0.5% and 2.0% of the real click rates.
That tells us the measuring tool itself works on real behaviour at a large scale. It does not tell us it works for career advice: that data is about clicks on a shopping site, not job applications, and it was a friendlier setting than ours.
What we haven't shown yet
- We haven’t run this test on Aethra’s own users. A read-only count of production data found too little to work with — a small number of accounts and close to zero completed analyses and applied suggestions. Our own backtest tooling refuses to report a result below 30 users and 100 decisions, on purpose.
- Nothing in the product today is “verified” or proven to improve your chances. The scores and improvement estimates Aethra shows are the AI’s own assessments, not measured results.
- The e-commerce result above shows the measuring method works on real logged behaviour at scale — it does not show it transfers to career recommendations. That setting was also friendlier to the method than ours: it ignored user context, and the underlying test still showed every option sometimes, where Aethra’s recommendations today are deterministic.
- When there is enough data, we’ll run the test and say what it found — including if checked advice turns out to make no difference.
What we're building next
We’re working on grounding the analysis in public job-market data — starting with O*NET, the U.S. Department of Labor’s occupation database — so a suggestion can point to the skills a role actually lists, rather than relying on the AI’s general knowledge alone. It’s still being tested and isn’t part of the analysis you get today.
Separately, we’re continuing to log resume analyses and the suggestions people apply, specifically so there’s eventually enough real data to run the same check above on Aethra’s own recommendations instead of a fashion-site dataset. No product behavior changes because of this — it’s passive logging of what already happens today.
Get in touch
Questions about the research, or interested in working with us? Email support@aethrahq.com. Buying for a team? Book a demo.