Projects & Competitions

K3Trash is a self-sorting trash can that uses a deep learning algorithm to predict the material of and drop an inserted item into the corresponding disposal bin. 

- Moves the objects using servo motors, stepper motors, and a linear actuator

- Utilizes a PIR sensor and load cells to detect the inserted item and trash bin weights 

- Coded in Python using a ResNet34 model, along with Arduino C++ code

- Connected to AWS cloud services through Wi-Fi to display trash data on a local network's webpage using Flask

- Won $1000 leading a team of 4 the semifinals of the UCI Sustainability Challenge

Write-up: https://www.instructables.com/DIY-MIDI-Drum-Kit/

This drum kit contains 7 piezoelectric sensors reading vibrations from each of the pads and a switch connected to an ESP32 microcontroller on a Protoboard. The microcontroller reads the sensor and outputs MIDI packets containing the note, channel, and velocity data from it. For a computer to utilize the packets, it needs to open a MIDI port and convert the serial output to a MIDI output on that port, as the microcontroller used does not have built-in MIDI communication support. The software also has debounce and threshold features to prevent unwanted hits or crosstalk between the different pads.

Led a group of 3 to develop an accelerated platform for the defect detection of a metal nut. A Raspberry Pi controls a camera to take an image of an object on a conveyor belt, sends it to an AMD Kria KV260 board for processing, receives a result back, and controls a servo to guide the object into the correct bin. The FPGA DPU on the KV260 speeds up common operations found in CNN models. As a result of this, we used a DeepLabV3 MobileNetV3 segmentation model to identify and localize scratch, bend, and color defects. This model was quantized to 8-bit int and compiled using the Vitis AI software stack.

SnapHealth is a service that aggregates patient health records from multiple different Electronic Health Record (EHR) companies for doctors and patients to easily access. Our group of 5 people competed in the Stella Zhang New Venture Competition,  an entrepreneurial process culminating in a "Shark Tank" style pitch, where we came 2nd place in our track and won $5000. I was the technical consultant responsible for creating a prototype app that gathered sample patient information from EPIC and Cerner APIs and organizing the JSONs into a readable and presentable format for the target audience.