A large staffing agency matches teachers and instructors with schools across Japan using a register of more than 50,000 candidates. Matching requires more than checking licenses and subjects: working conditions, experience, and individual preferences also matter.
Technica AI developed a way to structure inconsistent registration data and separate mandatory eligibility checks from flexible fit criteria. In project testing, approximately 90% of evaluated cases were reported as accurate. The figure describes test evaluations, not confirmed placement outcomes.
A large candidate pool, but complex placement requirements.
Staff needed to review teaching qualifications, subjects, school types, preferred working conditions, and experience to identify appropriate candidates for each opening.
Simple filters excluded some potentially suitable candidates and could not fully account for experience or preferences. Experienced coordinators therefore carried much of the decision-making burden.
Incomplete and inconsistent records limited matching quality.
Existing registration records included missing values, outdated information, inconsistent terminology, and free-text descriptions. Work history and current preferences were sometimes mixed together.
The business needed to make this existing information usable without replacing its long-established operating system. Coordinators needed a shared basis for comparison that made experience and preferences easier to assess.
Separating eligibility rules from suitability assessment.
Technica AI extracted relevant information from free-text and inconsistent records, converting candidate attributes and vacancy requirements into a more comparable structure.
Coordinators could distinguish mandatory conditions, such as licenses and subjects, from experience, working arrangements and preferences. Shared eligibility and scoring rules provided a consistent basis for comparing candidates, with AI organizing the information used in that assessment.
The approach produces candidate shortlists for staff review; it does not replace human hiring or placement decisions.
Approximately 90% accuracy reported in project test cases.
The team tested whether interpreted candidate information could support consistent assessments. Separating information extraction from score calculation addressed the inconsistent numerical results observed in early testing.
After separating AI-based extraction from deterministic scoring, the project reported that approximately 90% of evaluated test cases were accurate. The evaluation metric and sample size were not specified, and no post-deployment performance measure was reported.
The project made existing records more usable for candidate comparison. The next evaluation step is to confirm how well the shortlists serve current school requirements and coordinator workflows.
More consistent shortlists without replacing placement judgment.
-
Faster shortlist preparation
Structured information helps coordinators filter a large candidate database.
-
More consistent evaluations
Shared eligibility and fit criteria can help reduce avoidable differences between reviewers.
-
Better use of existing records
Free-text and inconsistent entries can become more usable for comparison.
-
Adaptable business criteria
Scoring rules can be revised when client or school requirements change.
Making Existing Candidate Records Useful for Placement Decisions.
The project gave coordinators a shared basis for comparing a large candidate pool without replacing the existing operating system. It organized experience, qualifications and preferences into decision materials, while keeping final placement decisions with people.