Definitely Not Bruno Mars

Implemented CycleGAN - a deep generative network used for voice conversion - using PyTorch to create a deepfake of an english-speaking male singer (Bruno Mars) to a korean-speaking female singer (IU).

I researched methods to improve CycleGAN using data augmentations.

I created a data pipeline that preprocesses audio waveforms into spectrograms using Librosa, trains and tests the model, and creates data visualizations in React (see left photo).

I reduced training time and memory footprint of the model by 50% by reducing CPU load during training time.

See the Presentation here
See the Results here

Meta

I worked at Meta in Summer 2023.

I was on the Instagram Reels RecSys Reliability Team.

I made DevX tools for my fellow engineers and worked on making IG Reels ✨reliable, fast, and efficient✨

Can't reveal too much because I signed an NDA xD

In other words, interview me for a role to know more about my experience ;)

but... if you can’t tell by how happy I am in this photo, I really enjoyed my time at Meta because of my great team. I also worked on really fun projects.

MicDrop

Senior Design project where I led a team of 4 in creating a karaoke video streaming service.

I researched many topics in system design for this project. Specifically, I used a CDN for streaming the videos I saved in a database

I created the entire software design, database design, and system design.

I also created algorithms for displaying text in sync with a video player.

*No demo at this time, sorry :(*