Saralin Zassman

Binocular Tension (2025)

A full-stack Python app that displays eyes looking in different directions based on the viewer’s position. It generates a point cloud in real time using data from a depth camera, then rotates it to correct for the camera’s height and angle. This correction allows the system to identify the person closest to the display and show the appropriate gaze.

Developed for media artist Rafael-Lozano Hemmer.

Python 100%

Repo

Global Level of Confidence (2025)

A full-stack web app that performs real-time face detection and landmark drawing using MediaPipe, and a Node.js backend that generates facial descriptors to find the closest matches from an archive of nearly 10,000 disappeared individuals during the Argentine dictatorship. The backend also uses a PostgreSQL database to store and update similarity scores over time, improving the accuracy of future searches. Visitors can explore their top matches and learn more about each person, helping raise awareness of this important historical period.

Developed for media artist Rafael-Lozano Hemmer.

HTML 87.3%, JavaScript 12.3%, Other: 0.4%

Repo

Recognition (2024)

A revamp of Recognition (2018) using modern, state-of-the-art technologies. Originally developed in openFrameworks, I ported the piece to a new stack with a Python frontend and a Node.js backend. While the concept appears simple, the piece posed significant technical challenges. It displays hundreds of videos simultaneously, performs facial comparisons, and continuously records and saves new footage to a growing database, all without introducing lag to the live camera feed or video grid, and running entirely on a single M2 Pro.

Developed for media artist Rafael-Lozano Hemmer.

Python 70.6%, JavaScript 28.6%, Shell: 0.8%

Repo

Transparency Display (2024)

An Electron app that detects body landmarks from a webcam and uses a custom segmentation pipeline to draw body figures. This approach allows finer control over visual parameters—such as limb thickness and length—rather than relying on off-the-shelf segmentation tools. The segmentations from each camera feed are composited, and pixel values are extracted and sent via DMX to drive the physical electronics.

Developed for media artist Rafael-Lozano Hemmer.

JavaScript 93.1%, CSS 4.4%

Repo

Mindful Scroll (2023)

A mobile application for mindfulness that encourages a slow and deliberate approach to colouring.

Master's thesis, accepted at ACM CHI 2024.

Note: CHI (ACM Conference on Human Factors in Computing Systems), is regarded as the top conference in Human Computer Interaction (HCI), ranking #1 on Google Scholar, with an average acceptance rate of 23%. Conference proceedings are the preferred avenue for publication in the field of HCI.

JavaScript 97.9%, CSS 1.4%, HTML: 0.6%

Repo

Speech to Text (2023)

Full-stack react app that converts speech to text using the Web Speech API, and then uses the text to generate a rhyme, poem, haiku and song using the OpenAI API.

JavaScript 83.6%, CSS 10.9%, HTML: 5.5%

Repo

Simple, rule-based chatbot (2021)

Uses Google's DialogFlow for response generation.

Created for a course project (CS 886 - Advanced Topics in Artificial Intelligence), to study the impact of user interface design on the human-chatbot interaction.

JavaScript 61.1%, CSS 26.1%, HTML 12.7%

Repo

DOM Data (2020)

Uses the Document Object Model of 100 webpages to generate structural priors to improve the vision-based segmentation of web pages.

Completed during my research internship with Dr. Michael Cormier at Mount Allison University.

Python 100%

Repo

Haiku Generator (2020)

Uses Markov chains to generate a large number of sentences from a collection of pre-existing poems. The resulting text undergoes rigorous testing to identify the optimal sequence of words that will be used in the final haiku.

Completed for a course project on computational creativity at the University of Kent.

Java 100%

Repo