Our February 2022 cohort just graduated a couple of weeks ago and we’re very proud to share the top final project from the Data Science program!
The final project is one of the key components of the Data Science program. In the last two months of the program, our students get to create a full data science project from scratch – from ideation to creating a model that actually works. This helps gain confidence as well as actually understand every part of a project’s lifecycle. Also, they write Medium articles and add the project to their portfolio!
Moodika: Generating Spotify Playlists based on Free Text Input
Project by Doron Reiffman, Noam Goldberg, Samuel Nataf
In this project, the team created a machine learning model which generates a playlist of songs from text prompts. The user enters text such as “driving on the highway” and the model selects relevant songs for the desired mood. This requires using NLP to understand the contents of the text, and which songs contain audio that best matches this content.
Read the full Medium article here
Multiclass Image Segmentation – Identifying Organs in Medical Scans
Project by Matia Cwajgenbaum, Maya Yanko, Richard Kampel
Radiation therapy is an important tool for battling cancer, but analyzing scans of organs is time-consuming and difficult for oncologists. In this project, the team created a model which identifies key internal organs in medical scans, savings doctors’ time and potentially making their work more efficient. This uses a deep learning-based architecture for image segmentation, taking medical images and identifying relevant regions within them.
Read the full Medium article here
Contract Review Classification
Project by Alon Gabay, Ursulla Smirnova, Yam Eitan
Reading through long legal documents is a challenge, and it is easy to miss important information. In this project, the team created an NLP machine learning model which reads through such documents and automatically extracts and classifies important clauses. The user can find this information at the click of a button without needing to laboriously search through the entire contact.
Read the full Medium article here