CHALLAPALLI KEERTHANA SAI LASYA PRIYA

Algorithmic Fairness in Student Performance (OULAD)

My project is about building an AI system that predicts whether students will pass or fail their courses, using a real dataset of 32,000 students from a UK university. The system is built in Python using machine learning models like XGBoost. The unique part of my project is that after building the prediction model, I also check whether the AI is being fair to all students - specifically checking if it’s treating disabled students and students from poor backgrounds unfairly. I then apply bias-correction techniques to fix any unfairness found, and use explainability tools to verify whether the fix actually worked. The system produces visual outputs like charts and comparison tables showing the model’s accuracy and fairness before and after correction. It’s a complete software implementation using Python, with a Jupyter notebook containing the full code, and a project report documenting everything

Credit Risk Fairness Audit

I built a credit risk model and then audited it for fairness - found the model was biased against younger applicants with a demographic parity gap of 0.22, and visualized everything in Power BI

AI vs Human Text (Ghost in the Machine)

Built a 3-class AI vs Human text classifier using DistilBERT fine-tuning and SHAP explainability. Designed a Genetic Algorithm adversarial attack achieving 84.3% human-confidence score exposing model vulnerability to stylistic perturbation

Student Stress ML

Built an ML pipeline predicting student academic performance and stress levels from behavioral data (login frequency, sleep, attendance). Presented findings at NCMLAI 2026, IIM Bangalore’s Analytics Society of India - first speaker in track.