Projects
Software Engineering Intern
IDX Exchange Property Search
Full-stack property search over 54,000 real MLS listings, with natural language search backed by a validated query builder.
- React 19
- Node.js
- Express
- MySQL 8
- Docker
- Jest
- Supertest
Justin Tan
MS CS student at Northwestern working on applied AI. I build systems that turn messy real-world input, video, camera frames, natural language, into something structured and usable.
Co-Founder & Founding Engineer
Pinpoint
Turns saved travel videos into map pins by extracting places from speech, captions, and on-screen text.
- React Native (Expo)
- Express
- Python
- Supabase
- OpenAI API
- Google Places
USC senior capstone, 5-person team
ChAIss
Android chess app that reads a physical board through the camera and plays or analyzes the position. I planned the project and chose the stack: Java calling Python through Chaquopy, python-chess for move generation, and a Roboflow-trained TFLite model for detection. Because detection isn't reliable enough to trust blindly, the board has a lock/unlock mode so a misread position can be corrected by hand and relocked. I built the UI, settings, and board theming.
- Java
- Android
- Chaquopy
- python-chess
- TFLite
- Roboflow
USC undergraduate research
Maritime Perception Pipeline
Computer vision pipeline for detecting and tracking small airborne objects against open sky, still in use at USC. I designed the pipeline structure around existing models with direction from my advisor: YOLOX detection, Real-ESRGAN super-resolution on the crops, ResNet-50 classification, and DeepSORT tracking, chosen over ByteTrack and SORT after comparing all three. Caught a train/validation leak caused by splitting video-sampled frames at the image level, which put near-identical frames on both sides of the split.
- Python
- PyTorch
- YOLOX
- ResNet-50
- DeepSORT
- Real-ESRGAN
Robotics coursework
Duckiebot Lane Following
Autonomous lane-following vehicle built on ROS and Docker, with a perception and control pipeline running on onboard camera input alone.
- ROS
- Docker
- Python
- OpenCV
Coursework
Audio Activity Recognition
ML pipeline classifying environmental audio events including clapping, coughing, alarms, and knocking, reaching 86% classification accuracy using MFCC, FFT, and RMS-based feature extraction.
- Python
- scikit-learn
- MFCC
- FFT
Embedded systems coursework
Nokia Ringtone Player
RTTTL tone generator with LED indicators on a Nucleo G474RE board, written in C++ and validated against LTSpice simulation.
- C++
- STM32
- LTSpice
Experience
Work
Education
Northwestern University
University of South Carolina
Beyond code
I've played violin since I was four. Fourteen years of private study, concertmaster of my high school and all-city orchestras, a Carnegie Hall honors performance in 2019, and roughly 28 solo recitals along the way. I still play: currently second violin in the Northwestern Philharmonia.
Also: soccer in high school, casual chess, and a Rubik's cube I solve slowly.