AI in Kasauti
AI that works beside your team, in the open.
Kasauti uses AI in two places: to test how candidates work with it, and to make recruiters faster. In both, the AI is visible, cites its evidence, and never decides.
For recruiters
What the copilot does.
Summaries against the rubric
Each application summarised against your must-haves, every claim quoting the CV line it rests on. Uncited claims are not shown.
Suggested scores
A score per rubric criterion with a short reason that quotes the answer. The reviewer accepts, changes or rejects it; the record shows who set the final value.
Drafts you send
Stage emails, rejection notices and interview invitations drafted from your templates. Nothing goes to a candidate until a person presses send.
Interview questions
Follow-up questions generated from the candidate's own test answers, so the interview checks that the understanding is theirs.
Job descriptions
A first draft from your criteria, with an inclusive-language check before you publish.
Search in plain language
Ask for candidates across your organisation in ordinary words, and get the people and the lines that match.
Question drafting
Scenario questions, seeded errors and rubrics drafted for a role, labelled as AI, for a named expert to approve or discard.
Hidden-text flags
Every CV is read by a deterministic parser that flags white, tiny, invisible or off-page text and wording aimed at a machine, so prompt injection is visible before any model sees it.
"Led the month-end close for a 40-entity group for six years, cutting close time from 12 to 6 days."
CV, page 1, Experience"Prepared VAT workings reviewed by the group tax lead."
CV, page 2, SkillsHow every output appears
Labelled, cited, editable, dismissable.
- Every AI output carries the AI label and a "why" link. Suggestions look different from decisions.
- Every summary and suggested score cites the exact parts of the CV or answer it relied on. Uncited claims are not shown.
- AI values and human values are stored separately, with who set the final one.
- The candidate-facing notice explains in plain language where AI is used, what it does, and that a person decides.
One door for every model call
The AI gateway.
Nothing in Kasauti calls a model directly. Every call goes through one gateway that enforces the rules in code, and the gateway is provider-neutral, so the model behind it can change without the product changing.
Minimise
Contact details and unneeded identifiers are stripped before a call. CV and answer text is passed as quoted data, never as instructions.
Pin
Model, prompt and rubric versions are registered, pinned and linked to the scores they produced. A change passes the same gates as code, plus the evaluation set and the bias tests.
Check
Outputs are checked against a schema before anyone sees them. Every call is logged. Cost and abuse limits apply per organisation and per candidate.
Keep it home
Region-pinned endpoints, providers with zero-retention terms and a signed data processing agreement, and no training on your data or your candidates' data, by contract and by configuration.
Bias testing and governance
Fairness you can show, not just claim.
- The scorer and the summariser are tested on paired examples that differ only in name, gender markers or nationality. Any difference in output blocks the release.
- Adverse impact is measured on outcomes at each stage wherever group data is lawfully and voluntarily collected, kept separate from the application and never shown to reviewers.
- The AI system register lists every AI feature, its purpose, model, data used, risks and owner, and is exportable for customers and regulators.
- A named owner reviews the regulations each quarter and updates the jurisdiction packs.
What AI never does here
Drawn in the product, enforced by tests that fail the build.
- Reject, rank out or advance a candidate. Status changes need a named person and a reason.
- Analyse faces, voices or emotions, or infer personality, health, religion, ethnicity, age, pregnancy or any other protected trait, including through proxies such as postcode or name.
- Use "AI text detectors" as evidence against a candidate.
- Scrape social media or buy data about candidates.
- Train on customer or candidate data.
Next step
See Kasauti on one of your own roles.
Bring a role you are hiring for. We walk through the hire end to end: the test, the evidence and the decision.