Job index

The retrieval corpus behind the agent: import roles and keep their vector embeddings up to date.

Roles in index

—

Roles vectorised

—

Dataset check (expected J0001–J0180)

Checking…

Import jobs from CSV

Columns are matched by header name. Recognised headers: job_id, title, company, location, work_mode, seniority, employment_type, track, required_skills, description, min_years. Rows are matched on job_id, so re-importing updates existing jobs.

Vector embeddings

Encodes each role into a 384-dimension embedding from its title, company, seniority, experience, location and skills — the vectors the agent searches at match time. The ground-truth track label is never included, so the benchmark stays leakage-free.