Trust & responsible AI

An AI that screens people owes you
an account of how it did it.

Talyn exists because résumé filters delete good candidates. A product built on that argument does not get to quietly rebuild the same harm with a better interface. Here is what the system does, what it refuses to do, and what we have not yet earned the right to claim.

01

Talyn does not reject anyone. It hands you the evidence.

The output is a ranked shortlist with a verdict line, not an automated hire/no-hire action. Nobody is auto-rejected, auto-advanced, or scored out of your pipeline without a person looking. That is the product design, not a setting — the machine takes the floor, a human takes the ceiling.

02

Every verdict opens into the exchanges it came from.

Each row carries the log: every escalation, every lifeline, the reason each one fired, and the candidate’s own words. If a decision is ever questioned — by a hiring manager, by a candidate, by a regulator — the basis for it is recorded rather than reconstructed.

03

No stylometric AI-writing detection. On purpose.

Detectors that judge writing style have high false-positive rates and misfire systematically on non-native English speakers and on anyone who writes formally. Talyn ships none. What it reports instead are process observations it can actually witness — an answer pasted in whole, an answer faster than a person can type, depth that collapses the moment the interview escalates on the candidate’s own claim — each with the numbers behind it. None of it feeds the verdict, and the candidate is never shown it.

04

Calibration examples are de-identified before they are stored.

When you calibrate a screen with an exemplar résumé, name, contact details, address, personal links, photograph, school and graduation year are stripped at ingest — before the row is written — and you can read back exactly what was kept. Those are the same fields a résumé filter discriminates on, and the same protected-class proxies NYC Local Law 144 and the EU AI Act scrutinise. Removing them is what stops “what a good candidate looks like” from decaying into “who our good candidates were.”

05

The interview is designed to find signal, not to trip people up.

A thin answer earns another angle at the same problem, phrased with the same warmth as an escalation, so nobody can tell which way they just moved. There is no one-way video, no timed gotcha, no personality inference, no facial or vocal analysis. A candidate has a conversation and does a piece of work.

06

What we have not earned the right to claim.

Talyn is not SOC 2 certified, has not completed an independent NYC Local Law 144 bias audit, and has not been through an EU AI Act conformity assessment. Those are attestations with dates on them, and we will list them here when they exist and not before. If your jurisdiction requires one of them for the way you intend to use this, tell us where you hire and we will tell you honestly whether we can support it yet.

Your obligations, briefly

Depending on where you hire, using an automated tool in a hiring process can require candidate notice, an accommodation route, retention of the decision basis, or a published bias audit. Talyn is built so those are possible — the evidence behind every verdict is retained and exportable, and no decision is fully automated — but the legal duty sits with you as the employer, and candidate notice is currently yours to give in your job posting or invitation. Ask us for what you need in writing and we will produce it.

Not legal advice.

Security question, data processing agreement, or something above that doesn’t hold up? hello@talyn.cloud.