Human Activity Classification Using Azure AutoML
Developed a data science pipeline using Azure ML, SQL, and machine learning models to classify human activity from sensor data, leveraging AutoML to achieve 94.1% accuracy with ensemble learning.
Programming
Data Science
Automated Cloud Pipeline for Antarctic Environmental Sensor Data
Automated a Google Cloud data pipeline using Python, SQL, Apps Script, Cloud Functions, and BigQuery to ingest, clean, store, and analyze environmental sensor data from Antarctica.
Geospatial Signal Strength Analysis Using BigQuery
Analyzed petabytes of vehicle telemetry data using SQL and BigQuery to map global modem performance, integrating Mapbox visualizations to uncover geospatial trends in cellular connectivity.
Geolocating Fuel Stations Using Vehicle Telematics and BigQuery
Built a scalable geospatial analytics pipeline using SQL, BigQuery, and DBSCAN to identify public gas stations from millions of vehicle fuel events while filtering false positives.
Machine Learning
Multi-Class Animal Classification with CNNs
Trained and refined a multi-class CNN classifier using validation tracking, Adam optimization, learning rate scheduling, and regularization techniques to improve animal image recognition performance.
Reinforcement Learning: Neural Policy Optimization in CartPole
Implemented reinforcement learning with a neural network policy using the REINFORCE algorithm to solve OpenAI Gym’s CartPole-v1 task through Monte Carlo returns and gradient-based optimization.
Real-Time Trail Segmentation Model for An Autonomous Robot
Implemented deep learning-based semantic segmentation in PyTorch to classify outdoor trails, benchmarking multiple CNN architectures and optimizing LEDNet for accurate real-time robotic perception.
Transformer-Based English-to-French Translation
Built a custom Transformer architecture for machine translation using deep learning and NLP techniques, implementing attention mechanisms, positional encoding, and decoding strategies for language generation.