Facebook AI Product Design Internship
AI data annotation platform design
In the summer of 2020, I joined Facebook AI's Tools & Platform Design team as a product design intern. Over the span of 12 weeks, I designed and launched a feature that improved AI data annotation quality for Facebook’s data annotation platform, as part of our team’s effort to ensure that AI models powering Facebook’s products operate at the optimal level. This project pushed me to explore and understand the AI problem space deeply while iterating on design solutions that meet the needs of different stakeholders and users.
While I cannot discuss the details of my project due to NDA, on a high level, I worked on researching the problem and similar platforms, aligning with my cross-functional partners to identify various constraints and requirements, exploring different design options, and conducting user testing sessions.
If you are interested in learning more about my project, feel free to reach out to me.
Product Design Intern, AI Tools & Platform, Facebook AI
Jun - Aug 2020
Figma, After Effects
Although I cannot discuss my project in depth, I can explain the general pipeline of building and launching AI models to demonstrate why it is important to ensure a high level of data annotation quality. The diagram above illustrates the stages of creating an AI model from scratch - collecting raw data, labeling them, training the model on the dataset, and deploying it for service.
You can think of the data labeling step as writing a textbook. When the label quality is bad, it is like teaching someone Math with a poorly-written textbook filled with mistakes. The Al model trained with these gibberish labels will produce much more inaccurate results after being deployed. Overall, poor data annotation quality would affect all the steps later in the pipeline, while high-quality labeled data could often improve a model’s performance. My project offered a solution to improve the quality of labeled data as a means to address this issue.
Always communicate and work with cross-functional partners
Each partner is an expert in their area. Frequent communications with them allowed me to constantly fine-tune my design directions, spot constraints early on, and move quickly through the problem space to bring out the biggest impact.
Keep scalability in mind
The more complex a problem space gets, the more costly one-off design patterns will become. Always design for scalability instead of tailoring a solution too much to solve only one edge case.
Complexity brings opportunity
AI, although very complex already, is still a rapidly developing technology with considerable potential. I am excited to design solutions for it to better work with people!