2026.10: Intent-aware candidate search preview, independent bias monitoring, expanded filtering and skill normalization in job search

AdeptID’s October 2026 release adds employer filtering and new custom filter types to Job Search, lets Job Search normalize skills in the same call, opens a private preview of intent-aware Candidate Search, launches a public dashboard for independent bias monitoring, and makes translation errors more descriptive.

Expanded Job Search filtering

What changed

New employer_name filter: Job Search now supports filtering on the employer that posted a job. Provide either one_of (keep only jobs from the listed employers) or not_one_of (remove jobs from the listed employers), with up to 100 employer names. Filtering is flexible: “Amazon” matches “Amazon Web Services.”

New custom filter types: Job Search custom filters now include integer, float, date, and boolean fields alongside the existing string fields. These match the filter types Candidate Search supports:

  • Integer and float fields (integer_1–integer_5, float_1–float_5) support min/max, one_of, and eq.
  • Date fields (date_1–date_5) support after/before, one_of, and eq.
  • Boolean fields (boolean_1–boolean_5) accept true or false.
  • String fields now also accept one_of (up to 100 values) and eq.

Example usage: Show a job seeker openings at specific employers by combining employer_name with one_of, or hide postings from specific employers with not_one_of.

Endpoints impacted

Existing

Action required

None. To use the new filters, add employer_name or the new custom filter fields to your Job Search requests.

Skill normalization in Job Search

What changed

One call instead of two: Job Search now accepts a new normalize_skills request field. When set to true, the candidate’s skills are normalized before the search runs. Previously, callers had to call normalize-skills first and pass the results to job-search, as the job search algorithm only recognizes normalized skills. Skills that can’t be mapped are not used, and certifications listed among the skills are treated as certifications.

Endpoints impacted

Existing

Action required

None. normalize_skills defaults to false, so existing integrations are unaffected.

Optional job ID for Candidate Search

What changed

New optional job_id request field: Candidate Search requests can now include a customer-provided identifier for the job being searched. The identifier does not need to match any job in AdeptID and is not used for matching. Sending the same value on every page of a search lets AdeptID group related requests, which improves support and search quality analysis.

Endpoints impacted

Existing

Action required

None. We recommend including job_id in Candidate Search requests and sending the same value on every page of a search.

Intent-aware Candidate Search (private preview)

What changed

Intent-aware search results: AdeptID's new Atlas model has been integrated into Candidate Search, providing recruiters with a more powerful search tool that considers how likely a candidate is to be interested in the job, in addition to their fit for it. This capability is available in private preview. It is not enabled by default, and standard Candidate Search behavior is unchanged.

Endpoints impacted

Existing

Action required

None. To learn more about the private preview, contact us.

More descriptive translation errors

What changed

Specific error codes: When a translation can’t be completed, the response now includes an error_code that explains why, plus a retryable flag. Previously, these failures all returned the same generic error.

  • 413 (input too large): translation_model_limit_exceeded, translation_output_limit_exceeded, or translation_input_too_large. These are not retryable. Inputs whose estimated output exceeds the service limit are rejected before translation is attempted. The response includes size details so callers can see how far over the limit the input is.
  • 502 (translation not completed): translation_timeout, translation_upstream_unavailable, translation_validation_failed, translation_invalid_response, or translation_upstream_failure. Timeouts and temporary provider failures can succeed on retry, though very large documents often time out again.

Inputs are never truncated, and partial translations are never returned.

Endpoints impacted

Existing

Action required

None. We recommend using error_code and retryable to decide whether to retry a failed translation. If your integration parses translation error messages, review the updated error responses.

Independent bias monitoring dashboard

What changed

Public bias monitoring dashboard: AdeptID’s Warden AI dashboard is now live. Warden AI is an independent AI assurance provider. The dashboard publishes Warden’s fairness monitoring results for AdeptID’s candidate matching and criteria assessment. Customers and their end users can view up-to-date bias testing results without requesting a report.

Endpoints impacted

Existing

Action required

None. Share the dashboard link with your compliance, procurement, or legal teams as needed.

AdeptID products are designed to support hiring and search (candidate sourcing, job search) decisions, not automate them. AdeptID systems are deployed within talent matching workflows where human judgment remains central to final outcomes. AdeptID believes that AI, especially when used in high-risk applications, must exceed high expectations for transparency, fairness, and accountability. This release has passed AdeptID's series of tests and controls for accuracy and fairness. See https://trust.adept-id.com/ for more information.