Hi,
I'm Justin Yang
Full-Stack Developer | AI Researcher
I build creative and innovative solutions.
I'm a recent graduate from Duke University with a major in Computer Science. As I value constant growth, I always seek to learn something new everyday, and I try to demonstrate the synergy of knowledge and passion in my work and in my personal life.
My main focus these days is building practical applications with friendly user interfaces that enhance productivity. I am a believer in well executed technology — we can make the most out of new, cutting-edge technologies when they are designed with the end user in mind.
When I'm not working in front of the computer, I'm usually playing my guitar, watching a movie, or looking for the next great spot for a scenic photo.
May 2024 - Present
Software Engineer
[Very Sweet Inc.]
Developed automated business budgeting feature for enhanced user engagement through Langchain Agent and tRPC API. Improved product utility by introducing personal tax optimization feature, identifying tax deductible transactions for each user.
Next.js
Node.js
PostgreSQL
Typescript
Langchain
Jan 2024 - Present
Research Assistant
[Col AI]
Implementing advanced methodologies to enhance the efficiency and extend the context length of Transformer models; Organizing internal meetings to devise and integrate diverse architectural frameworks, including RetNet, Mamba, and RoPE.
Python
Pytorch
May 2023 - Aug 2023
ML Researcher
[POZAlabs]
Performed invitational research on music generation task implementing fine-level dynamics. Conducted exploratory data analysis and observed success rate of 80% on valence mood classification task using Support Vector Machine. Accepted to AAAI 2023.
Python
Pytorch
Logic Pro
scikit-learn
May 2022 - Aug 2022
ML Engineering Intern
[POZAlabs]
Spearheaded ablation study for paper published in NeurIPS 2022 (ComMU) to investigate the effect of metadata in combinatorial music generation; evaluated its distinctiveness to human composed samples through Amazon Mturk involving 100+ composers.
Python
Jekyll
Pytorch
Box
Box is an automated blogging platform with custom CMS, OAuth, dynamic post management, and robust text and image generation. Implemented analytic insights to track user engagement and designed architecture with distinct public and protected services.
Next.js
Typescript
PostgreSQL
Langchain
Eventify Inbox
Created application to provide automatic schedule management service through email and calendar applications. Utilized LLM agent to create meetings and agendas in the users' calendar based on new emails received from Nylas API web-hook.
Flask
React.js
Langchain
Open AI
Earth Street Journal
Earth Street journal is a web-application that scrapes environmental news articles and summarizes them. Backend API was built using Flask and MongoDB to handle scraping, summarization, and user data. GPT 3.5-turbo model was fine-tuned with custom training and validation dataset to enhance data consistency and summarization quality.
Flask
Open AI
MongoDB
BS4
Bulk Buy Buddies
Bulk Buy Buddies is a web-application built to simplify bulk purchase decisions by matching Costco customers who wish to split purchased items. Backend was built with Flask and MongoDB to handle user, store location, product data, and product inventories. Frontend was built with Next.js.
Next.js
Flask
MongoDB
MID-FiLD
MID-FiLD is a research paper I worked on with the ML Team at POZalabs. MID-FiLD aims to enhance dynamics in music generation by using annotated data (cc#1 value in the modulation wheel). I conducted EDA and valence mood classification task using SVM. Paper is accepted to AAAI 2023.
Python
EDA
SVM
Not Uber
Not Uber is an academic project that simulates Manhattan using a graph-based model with over 50,000 nodes, 2,000 passengers, and 200 drivers to optimize driver-passenger matching. A variety of event-driven decision algorithms (A*, Dijkstra's, Floyd-Warshall) were used to minimize passenger wait and travel time and maximize driver profit.
Python
Contact
Connect with Me
If you want to know more about me, or if you would just like to say hello, send me a message. I'd love to hear from you!
