Justin Yu

Just a fun side project I made with friends.

We created a fully distributed terminal based arcade, which can automatically detect other computers on your network and connect to them to play games, fully inside the terminal.

The system is built on a custom implementation of the Raft consensus algorithm (dear god never do this again) for coordinating game state.

We’ve only implemented the game Tron so far.

Video Demo

Splash page for HackMIT 2020

“Amber alert” system for detecting Covid 19 cases in your area. Pulls up-to-date data from Johns Hopkins and New York Times and tracks and notifies you via sms alerts of new cases in your geographic area.

Demo

Stack: React/Flask + Twilio integration

This was a random side project done during winter break with some friends. Detects book covers and retrieves all book metadata in real time

This was my intern project during my summer interning at Dexai Robotics, a robotics food automation startup. I worked mainly on robot safety features, including a controlled emergency stop for the robot and designing a system of cameras for detecting and localizing humans in the workspace using MaskRCNN and computer vision.

Here’s a demo video of my intern project:

Here’s me experimenting with my detection system:

Fun fact: My intern project got a small shoutout in a video on the BBC :)

Motivation

HackMIT receives several thousand applications every year. As an organizer of HackMIT, I wanted to create a platform that would streamline the application reading process, automatically record and manage the scoring of these apps, and track the progress of indivduals throughout the app reading process

Tools Used

Frontend: Semantic UI and React
Backend: Flask server with PostgreSQL database

This was a project I worked on for HackRPI 2019.

Motivation

Companies today are collecting an unprecedented amount of data about their users - sometimes, a scary amount. We think that users should have a say in how their data is used. Since many companies have ambiguous opt-out systems that might even reset themselves without notification, we decided to take a novel approach to preserving user privacy. Tools Used

Built using Javascript and JQuery with Flask server backend for calling API endpoints

Description

Adsensible will automatically perform google searches in the background that are orthogonal in meaning to your actual google searches, thus obfuscating your real data with random noise. For example, for a random search, Adsensible may choose another word like “baker” to base noise searches off of. Then, it will generate realistic search queries using the noise topic performs the search. It also sends POST request to the tracking endpoints in the link so Google believes the link was clicked. This process is repeated 50 - 100 times, while remaining light on computational resources.

Description

Dormsp.am is an online platform that aggregates all events happening on MIT campus and displays them in an easy-to-use web interface. The goal is to change how students learn about and interact with events – rather than sifting through hundreds of emails, they will be able to see all events on a single calendar, filter by category, search by content, get matched with specific events based on personal interest etc. Tools Used

A script listens in on MIT dormspam (emails circulated to all students) and scrapes incoming emails. We then classify the emails as event or non-event, then extract key information from the event emails such as date, location and even the category of the event. This information is piped up to the front-end where it’s organized into a calendar and will display based on user preferences.

Currently in Progresss

Live Demo

Stack: React / Flask

MIT Infolab

2018 - Present

Current undergraduate researcher at MIT CSAIL, pursuing projects in improving question-answering systems through NLP models.

The Infolab

The Infolab is one of the labs located in MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL). The primary work of the Infolab is to investigate “natural language and visual perception in different contexts from how it is learned, to how it is grounded in visual perception, all the way to how machines can readily interact with humans.” Many of the projects utilize a variety of natural language processing and computer vision tools, with a unique desire to combine classical AI and machine learning.

Current Research Work

My current research project focuses on using recently developed sequence models and other machine learning tools to improve the performance of question-answering systems, such as Infolab’s START system. Much of the information retrieved from knowledge bases such as Wikipedia can be stored in triples of the format (Subject, Relationship, Object).

My task is to develop various sequence models to generate natural language questions given a triple of information i.e. (Obama, Spouse, Michelle) → “Obama’s wife is Michelle”. With these automatically generated questions, START will be able to better answer real human questions about these subjects.

I also built a web platform for creating written annotations for educational videos. These annotations are fed into START’s knowledge base, along with the transcript of the video, so that users are able to ask context questions about specific videos. The hope of this project is to apply automated question answering systems to MOOCs and educational platforms.

Botball Robotics Team

Demo of a robot I worked on

Botball 2015 Competition Video

I was a Captain of my high school’s Botball robotics team. I probably spent more time at Botball than any classes I ever took in high school.

Botball is a robotics competition for high school students. Unlike competitions such as FIRST or Vex, Botball is a fully autonomous robotics competition, meaning there is much more emphasis on software and designing robots that are robust given hardware limitations. Since there is no human controller, robots must have built in recovery mechanisms, as even the smallest of errors can send the robot spiraling out of control.

The robots are given a wide variety of tasks, such as collecting certain game pieces or sorting items by color. You can check out a practice run of one of the main robots I worked on here.

While I was on the team, we won:
1st place in New England Regional Competition 2015, 2016, 2018
1st place in Global Competition, 2015
3rd place in Global Competition 2016
6th place in Global Competition 2018

Botball was my first endeavor into robotics, and I learned and grew a lot from the experience.

Based in the Boston area, MAHacks is a high school hackathon I founded in 2016

I wrote down some of my motivations for starting MAHacks here. You can also check out pictures from previous events here

In 2016, disappointed by the small number of hackathons available for high school students in the Boston area at the time, I decided to found my own high school hackathon with a group of friends.

MAHacks was truly a learning experience for me and tested my creativity, organizational and team management skills. As director, I spent countless hours meeting with sponsors, reaching out to mentors, holding team meetings and delegating tasks. I worked to make sure MAHacks was enjoyable, educational, and an event that would introduce newcomers to the wonderful world of hackathons.

Over the course of two years, we grew from ~60 attendees to over 100 by the time I graduated and passed on the directorship.

I worked as a software engineering intern at Accion Systems during my high school junior summer.

Accion Systems is a startup that produces ion thruster chip technology for micro-satellite propulsion. I was lucky enough to work as an intern here during my junior summer of high school.

The specifics of my project are under NDA, but as an overview, I helped Accion build low-level software that would allow them to reprogram their thruster chips floating in orbit from the ground. I also improved the custom data processing GUI they used to analyze testing data.

During my internship, I had to learn a lot on the fly: microcontrollers, UART, assembly, as well as familiarizing myself with python scientific librares (numpy, pandas) and GUI libraries (tKinter)

In the summer of 2016, 45 students from around the country were invited to participate in BWSI, a program to teach advanced autonomous robotics to high school students.

MIT Beaverworks Summer Institute was a residential summer program at MIT where we learned the fundamentals of robotics, control theory, planning algorithms and more.

Technical presentation of what I learned at Beaverworks

Throughout the program, we:

• Completed various technical challenges, worked with the RACECAR platform, using LIDAR, camera and IMU sensor input.
• Learned to use software such as ROS and Gazebo to program more complex robots.
• Implemented a PID controller, a potential field path planning algorithm and SLAM.
• Used computer vision to detect and recognize colors and objects.

The four weeks culiminated in a final race between all teams, where my team won 2nd place.

MIT Beaverworks Summer Institute was amazing, and gave me experience working with the kinds of sensors and hardware that would be implemented in the real robotics industry. It also heightened my interest in robotics and technology.

In 6th grade, a group of friends and I decided to create a “Tacit Glove” to help a blind student in our grade navigate the halls.

The “Tacit Glove” was inspired by something we saw in an issue of MAKE magazine that we wanted to recreate. It’s a device that is worn on the hand equipped with ultrasonic sensors in the front and motors in the back for haptic feedback when navigating an environment.

We worked on it for a year and presented it to our blind friend. Though not a perfect solution, it was exciting to be able to create technology that could help make someone’s life easier. This was my first real technology-related project, and it inspired me to pursue my passion for using technology to solve important problems.

Later, we expanded the idea to using wearable technologies like Google Glass as a platform for helping the visually impaired navigate, leading me and my friends to found Auxivision.