Problem
Given a person's profile and a target role, how far are they from being job-ready — and which gaps matter most? A single yes/no readiness label hides the severity.
Data
Profile + role data transformed into an 82-feature engineered dataset covering skills, experience and education signals. Full ingestion and transformation handled in-pipeline. TODO: rows / source / label definition.
Approach
Trained and compared multiple model families on the same features and split, then wrapped the best one in a Streamlit app that does ingestion, transformation and real-time inference from user input.
Outcome
A working app that takes a profile and returns a severity score with the driving gaps, not just a binary label. TODO: chosen model + headline metric.
Limitations / next
- TODO: label subjectivity / ground-truth source
- TODO: generalisation beyond the training population