What I’ve learned in Meta

As a data science intern in Meta, I learned how to navigate myself in big companies, appreciate the engineering mindset in the tech industry, and move fast through project planning.

I worked in Metaverse.

Overall learning

• Stalk the data owner: Data in big tech are usually processed through many teams and have been customized to accommodate different needs. I analyzed the wrong dataset initially because I didn’t talk to the original data owner. Before jumping into analysis, I learned to search the dataset and relevant work of data owners in the workplace.
• Build a mental model for engineering: I hated engineering because debugging felt like shooting in the dark. When I was agonizing over building up a data pipeline, I noticed that my manager could figure it out quickly, even though he didn’t have experience. Why? I learned that he had a much better mental model for data pipelines because he had read many technical books. After I picked up “design a data-intensive application” he suggested, I began to learn how machines handled databases and distributed computation systems. My confusion (i.e., concurrency and partition) stopped being a myth, and I noticed I had improved in formulating better bug hypotheses, and my debugging experience became less rocky.
• Be pushy at project planning: For planning better, I learn that planning doesn’t only include what I plan to do but also what I plan to achieve and how to do it. The template for me is: To do & Expected output, so you have something measurable by the end of every day.

• What can we do if the plan doesn’t go well in the middle? Usually, I tried to work harder and set more aggressive goals, hoping to get things on track. Those plans were vague and not valued because I was too busy to plan, miscalculated my productivity, and mentally stuck with the initial plans. Thanks to my manager, he would tell me though he was disappointed we weren’t on track, my plan wasn’t realistic at all. He would push me to re-plan a new schedule in a detailed manner (action & expected output), ensuring I could and will follow my new plans. I felt less stressed and more confident in achieving them. I learned that adaption isn’t about taking lots of pressure but being a good planner, learning to be clear-headed from unexpected situations, and continuously pushing myself to re-plan realistic goals.
• Hypothesis-driven planning: If the plan is data science-related, I will write down hypothesis and testing plans before digging into the analysis. For instance, if we ask why our singer dataset is unreasonably big, I will start by writing down several hypotheses (e.g., lots of duplicates, low popularity singers, or names that aren’t even singers). To test the same hypothesis, I then, therefore, break down four execution steps 1) propose several candidate algorithms to define duplicates for computers, 2) recommend the best candidate based on their potential performance(accuracy vs. recall trade-off) and computational resources needed. 3) write down simple toy examples to demonstrate how the algorithms will work (e.g., Apple, apples, Apple tree, app league), 4) brainstorm potential edge cases to break the algorithms (i.e., when A = B ≠ C is A, C duplicates?) The purpose of writing them down is to prevent me from digging around random insights during the process. I saved time and was more directed toward my goal.

Personal interests

• Start a “weekend project”: Whenever I asked my manager, “how’s your weekend?” He always said, “Coding is my side project.” A few weeks after those answers, he got me curious. What was so fun that he wanted to spend his weekends on it? I stalked his GitHub and found lots of astounding websites. He said those were his practice websites and he built them over a weekend. A weekend? Looking at my engineering friends, they all started suffering for months for ugly projects. How did he finish over a weekend for such a beautiful website? I later learned that he found good templates to imitate rather than learning and building from scratch. There are two benefits. First, you know the correct answer so that you won’t get stuck for too long. Second, you are promised to have a decent outcome. That’s why he was able to learn fast while feeling fulfilled. Since it’s time-bound and attainable, his idea of a weekend project motivated me to start an engineering project.

I apologize that my stories sounded quite personal and did not present the overall Meta. Yet, in a big company, none of the experience is generalizable, and learning is always a personal journey. ☺️