Two thousand AI-written applications, or one page you can inspect

Every claim on a profile says how it's backed: stated, linked to a source you can open, or confirmed by a collaborator or issuer. A file the candidate uploaded is still theirs — only the top labels are independent, and those can't be self-selected. Ask a question on the page and get an answer grounded in that evidence, with citations. If the evidence isn't there, the answer says so instead of guessing.

Questions asked on a profile are visible to the candidate. That transparency is the product working as designed: AI hiring rules now carry notice duties (Illinois, today) and record-keeping and human-review duties (Colorado from 2027, the EU's high-risk regime from late 2027). Your workspace can export the full Q&A audit log as CSV — timestamped, attributed, citation-counted — whenever those records are asked for. A tool that shows its work is the tool you can defend. How matching works is public, verbatim.

There are no scores, no rankings, and no automated rejection here — matching is deterministic rule-checking on stated requirements and stated terms, and it yields an unranked qualified pool. You invite; the candidate accepts; the conversation starts with their evidence already on the table. Postings disclose compensation, so neither side wastes a round.