Built a BERT-based multi-label emotion classifier to detect 7 emotions from English text. Used the bert-base-uncased model with frozen embeddings for efficient training. Designed a Keras model with Bi-GRUs and dropout layers. Trained using TensorFlow with mixed precision and evaluation based on micro F1-score. Built a complete pipeline using Hugging Face libraries.

Built an IoT-enabled web app to monitor heart rate and oxygen levels in real-time using MAX30102 sensors. Backend used Node.js, Express, MongoDB, and REST APIs. React frontend displayed charts. Synchronous state machines captured data. Offline storage synced automatically on reconnection. Configurable intervals allowed flexible data capture.

Built a Random Forest Classifier to predict ODI match outcomes with 87% accuracy. Developed a real-time dashboard that updated predictions every 5 minutes using simulated data. Backend APIs delivered dynamic stats. System improved fan engagement through evolving match outcome predictions.

Developed a dynamic dashboard to visualize cricket stats from 150,000+ records. Simulated live matches using backend APIs connected to a Quarto-based frontend. Delivered insights with bar, line, donut, and map plots. Built APIs for batting, bowling, and partnership analysis with real-time updates.

Analyzed traits of 890 fictional characters across 100 universes using psychometric and MBTI data. Cleaned and merged datasets, performed correlation analysis, and created visualizations using bar, radar, and density plots. Revealed insights into character likability and audience perception.

Built a surveillance system using CNN-based mask detection (99% accuracy), social distance monitoring, and temperature sensing. Implemented socket programming for multi-node data transmission. Used Flask and SQLite for lightweight front-end deployment and data storage. Published in IJRASET.