Skin Cancer Recognition
I developed a custom convolutional neural network (CNN) to classify skin cancer images, using a Nature Publication as inspiration. I trained the CNN I developed on more than 10,000 images of various types of skin cancer lesions. To promote model transparency, I leveraged Gradient-weighted Class Activation Mapping (Grad-CAM) to further validate the model's predictions. I utilized an Nvidia L4 GPU and CUDA on AWS using a g6.xlarge EC2 instance and a Jupyter Lab server to accelerate training and inference.
Spotify Stats
I developed an application using several AWS services to showcase my top tracks and artists based on Spotify streaming data. The application seamlessly retrieves my streaming data from the Spotify API every 30 minutes and logs it as a CSV file in an S3 bucket.
At 12:01 AM daily, two Lambda functions are triggered via EventBridge. The application uses four DynamoDB tables in total: two for tracks and two for artists. One set of tables stores the current day’s top tracks and artists, while the other set stores the previous day’s top tracks and artists. Maintaining these two sets of tables allows me to track changes in stream counts for my top tracks and artists.
To visualize ranking changes, I added green arrows to highlight when a track or artist surpasses another based on stream counts. Additionally, I calculate and display any increase in streams from the previous day to the current day for each top track and artist. The data is retrieved from the respective DynamoDB tables for top tracks and artists.
My top tracks and artists are on my website and are updated daily.
Spotify Wrapped
I recreated Spotify Wrapped using my lifetime Spotify streaming history.
Spotify Grid
Using data from my Spotify Wrapped project, I knew my top songs, albums, and songs. With this data, I created an image grid of my top album covers and artist images with web scraping and the Spotify API.
Spotify Recommendation Engine
Using over 1000 of my top streamed songs from my Spotify Wrapped project, I used a decision tree classification model to predict new songs based on my top song's musical attributes. I used the Spotify API to create and publish a new playlist entirely created from my model.
NBA Efficiency
Using the NBA's offensive and defensive team ratings, I plotted each team by their offensive and defensive rating. I normalized each team's offensive rating and defensive rating and used their logo to place them on a net rating chart.