Problem
Manual CV intake and screening was slow and inconsistent — roughly six minutes of human effort per application before a recruiter could even compare candidates.
Data
Unstructured CVs (PDF / DOCX) of varying layout. Built a robust ETL step for intake, parsing and field-level validation before any scoring. TODO: dataset size / source.
Approach
An autonomous agent orchestrated with UiPath and Python drives a multi-stage pipeline: extract → normalise → validate → score → route. Each stage is independently checkpointed so a bad document fails loudly instead of silently corrupting downstream steps.
Outcome
Processing time per CV dropped from ~6 minutes to 4.4 seconds, turning a manual bottleneck into a background job. TODO: accuracy vs. human screening.
Limitations / next
- TODO: parsing failure modes on unusual CV layouts
- TODO: bias / fairness checks on the scoring step