PROJECT
Bias Detection and Mitigation in Pre-trained Language Models
Developed a bias detection and mitigation framework using PyTorch and Hugging Face Transformers, analyzing over 17,000 instances from StereoSet and 1,508 sentence pairs from CrowS-Pairs to achieve fairness in gender and profession with minimal accuracy loss. Led a team to design an adversarial training framework for debiasing language models, addressing societal biases in recruitment and healthcare, and deployed a scalable debiasing pipeline on Google Colab.
Mobile Price Prediction System using Machine Learning Algorithms
To enhance mobile phone price forecasting, I developed a predictive model using Python and Scikit-learn, analyzing 1,000 instances with 19 features to achieve 88% accuracy. By conducting data analysis with Python and SQL, I identified key pricing factors, improving prediction accuracy by 12%. I then deployed a Flask web application to provide real-time price predictions, supporting over 500 daily queries and reducing user decision time by 30%, demonstrating my ability to leverage data science and web development skills to deliver practical, user-friendly solutions.
CD/DVD Management System
To streamline media collection management, I developed a CD/DVD library system using SQL, RDBMS, and Java. By implementing core RDBMS concepts like DDL and DML operations, I reduced database response time by 30%. I designed a structured database with primary and foreign key relationships to ensure data integrity and enabled CRUD operations via Java and JDBC. The system features an interactive UI with search, categorization, and tracking functionalities, along with loan management capabilities, including overdue item reports and member reminders. This project highlights my expertise in database management and software development to deliver efficient, user-centric solutions.
Cyberbullying Detection System using Naive Bayes Algorithm
Developed a high-accuracy (90%) cyberbullying detection system by implementing a Naive Bayes classifier with N-grams to identify abusive language patterns, contributing to safer social media interactions.