Awarded 1st place in IEEE Competition at UC San Diego

Leveraging the AMG8833 IR thermal imaging sensor, my team and I manufactured a prototype that detects human presence within 7 feet of the user. The sensor is a 8 x 8 pixelated thermal imaging camera that produces a "heat map."

Once we had the sensor running with an Arduino Uno board, we collected thermal imaging data in various distance and situations: without a person, with a person, with multiple people. Upon data analysis, I found out a temperature threshold at 7 feet distance and learned that human(s) are detected as columns in the thermal image. I wrote up the code for human detection using this fact in Arduino C and went through validation. It was a huge success and I had a lot of fun merging software and simple hardware.

---

As a team lead of 10 computer science major, we developed a local-first web application using agile methodologies. For the 8 weeks of development, our team was able to work through 3 sprints, each followed by sprint retrospective meetings. In terms of my individual contribution, I architected the CI/CD pipeline for deployment, allowing 5 releases, each with iterative development. The components of the pipeline include ESLint, JSDocs, and Jest Unit Testing. Furthermore, I did front-end work using HTML and CSS to create and design character and game pages.

The biggest take aways from this project were documentation and agile methodologies. We created Architectural Decision Records (ADRs) to reflect upon our development choices in the future. In terms of agile methodologies, I learned the importance of iterative development and how to bring a team together for collaboration.

---

This was a class project in which my team and I fine-tuned BERT encoder to improve its test acurracy on Amazon Massive Intent Dataset from Amazon Massive Intent Dataset. We elected methods of layer-wise learning rate decay and scheduler with warm-up steps to achieve 86.1% accuracy.