How JAI Bot uses AI

What the AI does

• Extracts job requirements from the employer’s job description. The employer reviews and approves them before any candidate is graded.
• Compares each application with those approved requirements, citing the exact resume lines it relied on, and rates each requirement on a 0–4 evidence scale: direct evidence, transferable evidence, or not documented.
• Optionally reviews interview transcripts and public professional sources for the same job-related requirements, only with the notice and consent described to the candidate.
• Suggests interview questions to clarify anything that is not documented.

What the AI does not do

• It does not make hiring decisions. A person at the employer decides, and JAI Bot never auto-rejects an applicant.
• It does not use or infer protected characteristics such as race, sex, age, disability, religion or national origin. Voluntary demographic answers are stored separately and never shown to the AI or to reviewers.
• It does not analyze faces, voices, emotions, accents, tone, fluency or personality, and it does not estimate honesty.
• It does not score “culture fit”, predict retention, or penalize employment gaps.
• It does not treat a missing keyword as proof someone lacks a skill: undocumented items are shown as questions to ask.

Safeguards

• Notice before applying: every applicant is told how their application is reviewed and can ask for a human review or an accommodation.
• Documents are data, not instructions: hidden text or embedded “prompt injection” in a resume is detected and sent to human review instead of being graded.
• Consistent standards: all applicants for a job are graded against the same approved, source-linked requirements and weights.
• Authenticity checks are prompts, not verdicts: signals such as text copied from the job posting route to a person for verification and never change the score.
• Fallback transparency: if AI grading is unavailable, a clearly labelled provisional score from a keyword and date rules engine is shown and the AI grade is retried.
• Monitoring: employers can review selection and scoring rates by group (from voluntary self-identification) and export the data for an independent bias audit.
• Records: employers control retention, and candidates can request deletion of their JAI Bot profile.

Models and data

JAI Bot runs on secure cloud infrastructure. Grading starts with JAI Bot’s built-in engine, which checks factual requirements (years of experience, degrees, licenses, tools and practices) directly against the resume’s dates and wording without AI, and publishes a grade only for the kinds of checks that have tested as highly reliable. The remaining requirements are reviewed by large language models from established AI providers — must-have criteria are reviewed independently by more than one model, and disagreements are flagged for a person — with a rules-based fallback. The AI does not see the candidate’s name, email, phone number, profile links or street address. JAI Bot also re-reads each resume in a different order to catch grades that depend on text position, requires specific details (numbers, dates, named tools or scope) before giving top marks, and keeps a tamper-evident record of every assessment. Candidate documents are used to provide the service to the employer the candidate applied to; we do not use them to train third-party AI models. Payments are processed by Stripe.

How the grader improves

JAI Bot’s built-in grader learns from de-identified grading results across employers, never from who was interviewed or hired. It measures how often each kind of built-in decision matched the final grade and sends less reliable kinds back to review, and it learns general work wording that reviewers at several employers confirmed shows a skill. Learned wording can only give credit for work a resume describes; it never lowers a grade. The built-in evidence score (0–100 for each requirement) also uses a small text-similarity model to find the resume line closest in meaning to each requirement, so work described in different words still counts; it does not write text or make decisions, and lines that are only names or contact details, or that mention protected traits, are left out of that comparison. The score learns how much each job-related signal matters, such as key words found, real work described, numbers and results, recent roles and years asked for, by comparing its scores with full reports and with employers’ own reviews; a change is used only when it is measurably closer to those grades on results it did not learn from. Each change is tested before it affects any score, must give identical grades when only protected details (such as name, gender, race, age, religion, disability or veteran status) change, and is monitored with the four-fifths guideline on voluntary self-identification data in aggregate. Changes are logged and can be paused at any time. Learning never uses names, contact details, employers, schools, places, dates, self-identification answers or hiring decisions. Details are in our Privacy Policy.

Ask JAI, the workspace assistant

Employers can ask the assistant in their workspace to explain how JAI Bot works, to find resume lines about a skill, and to list work waiting on them, such as candidates waiting for interview feedback. It runs inside JAI Bot without an outside AI service and reads only the employer’s own workspace.

• Quotes, not verdicts: every match shows the resume lines it is based on and the job they appear under. Matches are found by wording, so the employer reads each quote before relying on it.
• Missing is not a no: when a resume doesn’t mention something, the assistant says “not mentioned”. That is treated as unknown and never counted against the applicant.
• Confirmed gaps need a record: a gap is shown only when a person on the employer’s team reviewed that requirement and marked it not met, or when the resume’s own dates fall short of a years requirement.
• No searches on personal characteristics: searches for age, sex, race, religion, national origin, disability, pregnancy, family status or similar are refused.
• Recorded: each resume search is listed in the workspace’s activity log.

For employers

Laws on AI in hiring differ by location and are changing, including New York City Local Law 144 (bias audits and notice), Illinois requirements on AI notice in employment decisions, the Colorado AI Act, and the EU AI Act. JAI Bot provides notices, human-review controls, and audit data to support compliance, but each employer is responsible for its own legal obligations. Consult employment counsel for your jurisdictions.

Questions or requests

Job seekers can ask the employer for a human review or accommodation, or contact contact@jaibot.ai. See also the Help Center.