• I built an Airtable-like product and then a personal knowledge system.
• However, I stopped working on this product because I don't believe these approaches are effective in solving people's problems.
• Technologies: Typescript, React, Stitches, Radix-UI, TailwindCSS

A technical leader and a multi-hat hat person. I took part in every aspect of building Umbala.Tv, a startup about social e-commerce platform.

My notable works while working at Umbala include:

• Led a team of five in the design, development, and deployment of a scalable video streaming platform to serve streaming for Umbala.tv app. The system was built from scratch and can deliver low-latency streaming (<3s) with peer-to-peer (P2P) features to cut CDN load up to 60%. Early demo: youtu.be/oeAap6oJ4Uo
• Built a p2p streaming protocol (it's like Torrent but for live stream). POC: youtu.be/GGVW60x54jE
• Designed and implemented APIs, webapp, admin dashboard,... for Umbala.Tv
• Technologies: Golang, Postgres, Redis, React

• Built and scaled Admicro's advertising system, which handles more than 600M ad requests per day, making it one of the largest ad systems in Vietnam.
• Led a team of engineers responsible for building ad services and many data systems to handle all things behind the scene.
• Worked on some interesting projects for advertising system such as ip2location, DMP, frequency capping, dynamic ads recommendation system, cross-device tracking, ...
• Technologies: Hadoop, Apache Spark, Grpc, Redis, Cassandra,...

• Researched and developed web games using HTML5 technologies. I developed 5 games using different technologies such as Raphaeljs, Box2D.js, Tween.js, Three.js,..

• A Linear App clone with React and Tailwind CSS. An experiment with Tailwind CSS.

• One-week side project to learn about TailwindCSS. It was a top post on ReactJS's reddit. I wrote about my experience here: reddit.com/r/reactjs/comm…

• Implementation of Spleeter, a popular music source separation library, in Pytorch. It can load 2-stem model to separate voice from instruments.

• An experimental CNN based neural vocoder, which generates speech from Spectrogram. It's in my attempt to build much faster vocoder than WaveNet. It was inspired by m-cnn architecture.

• A visual search engine like Pinterest, based on Elasticsearch and Tensorflow, which may be one of the first open source Deep Learning-based visual search engines.

• I built this in two weeks of hacking, trying to build a visual search based on what I had learned.

• This project was featured on Github's Trending.