Machine learning · team project · 2026
Placify AI
Placement data, out of the spreadsheets.
A placement-prediction platform for colleges: a CatBoost model over 5,000+ student records, dashboards for coordinators and a résumé parser.
84.7%
accuracy
94.5%
ROC-AUC
5,000+
student records
3
people on the team
The problem
Placement data was scattered across spreadsheets, résumés and manual updates from coordinators.
There was no clean way to see who was ready, track a cohort or tell a student what to work on next.
How it works
Step 1
A model coordinators can trust
A CatBoost classifier trained on 5,000+ student records predicts placement outcomes, and was compared against SVM and logistic-regression baselines.
Step 2
Structured data first
ETL workflows clean and normalise the records into PostgreSQL, which feeds KPI and cohort dashboards built in React.
Step 3
Résumé parsing as a service
A FastAPI service parses résumés and runs a skill-gap analysis, so the advice is specific to the student.
The hard parts
- Messy inputs
- Records came from inconsistent sources, and résumés arrive as untidy PDFs with free-form skill lists.
- Beating simple baselines honestly
- The model had to outperform SVM and logistic regression while staying explainable enough for coordinators to act on.
The proof
- 84.7% accuracy and 94.5% ROC-AUC, ahead of the SVM and logistic-regression baselines.
- Dashboards and structured APIs in place of manual sheets.
Honest statusPlacify AI is a team project built with two classmates.