Dickson Tsai

April 2025 - Present: Product engineer on Claude Code March 2024 - April 2025: Research engineer on the Tokens team.

Tech lead of Search Experience Growth Engineering

Tech lead of Image Search Metrics Engineering

Taught SAT Math and ACT Math/Science classes, ranging from 5-20 students. Taught private tutoring sessions for Computer Science (Python, Java), SAT Reading, Algebra 2, and Pre-Calculus.

Worked on the Ground Truth team to build great learning datasets for self-driving cars.

Full stack data analysis engineer in Search Growth and Analysis, focusing on logging in the Android Google app. Contributed to client logging infrastructure (e.g. configurable client-side log sampling), log format standards, pipeline management (e.g. hourly data aggregation), dashboarding, logs quality monitoring, and one-off analyses for experiments/releases.

Tools for Cloud Services Localization/Siri Language Engineering, the team responsible for localizing Siri.

Summer 2014, Fall 2014, Spring 2015 - CS61A (Structure and Interpretation of Computer Programs) Introductory course for CS majors covering functional, object-oriented, and declarative programming; basic recursive data structures; interpreting programs just as other data. Taught in Python. Accomplishments: - Guided a group of 30 students through labs and discussion. - Contributed to office hours queue web application. - SQL aggregation lab focusing on linguistic data as an application. What worked: - Deconstructing the process for drawing simple environment diagrams. Make sure students have mastered every single step. This can take several iterations of examples, exercise, and sharing. - 1-on-1 appointments with students. - Physical demonstrations. Had students count the number of people in the room using tree recursion. - Big picture. Share experiences, applications to other domains. - Focus on teaching problem solving, not go through the work packets. - Examples. Concepts are the end, not the means to the end. What didn't work: - Teaching while overburdened. Sleep is important. - Saying "___ is easy" to reassure students that they can do it. It had the opposite effect. - Going through queue too fast in crowded office hours. Only identified bugs and then asked students to "think about it". Got students unstuck, but did not do as much for understanding. - Explaining using rules/abstract concepts. Examples and worked better. - Giving students too little to work with. Examples were more engaging, not just hints directly related to the solution. - Focusing on perfection. Recognize you're not perfect either. Focus on variety of approaches and creativity. - Working remotely. Pair program with fellow TAs for more creativity/feedback. - Not disagreeing constructively. Following meekly and reflecting after the fact does not help anyone.

Provided feedback to student homework submissions and made video solutions on homework/exam problems. What worked: - Taking the initiative to carry out administrative tasks for your project, so that your managers don't have to. - Automating as much as possible using scripts What didn't work: - Failed to check in regularly with head GSI and professor to talk about workload, falling behind. - Thought too much about providing perfect, thorough feedback, which was not necessary at all. Timely feedback was much more important. - Not dedicating chunks of time to focus in particular on grading. Being intimidated instead on the work that eventually had to be completed, which delayed actually working on it.

http://linguistics.berkeley.edu/~csheil/gaelic Accomplishments: - Built a searchable MySQL database of Scottish Gaelic sentences/morphemes - Angular.js frontend, REST backend in PHP Slim - Implemented authentication system for registered users to edit sentences/chapter metadata, to create personal collections of sentences for more efficient research What worked: - Previously tried Python Tkinter interface, but web application was a much better choice of implementation. Easier to control interface visuals and standardize data into a specific schema. - Thinking in terms of what features are necessary rather than how the app should be implemented is effective at getting the ball rolling. What didn't work: - Project fell behind (actually completed over Winter Break 2014-2015), because I had trouble dedicating time consistently to the project. - Should have planned out project maintenance and collaboration.

Graded and gave feedback on students' style ("composition") for each of their projects written in Python. What worked: - Explaining the purpose of concepts: assert, helper functions, dictionaries to store intermediate computations instead of just suggesting edits. - Providing a well-written summary of feedback, not just a number. - Showing students how to tie course concepts like inheritance to clean up project code, because students don't have exposure to that perspective. - Providing references to students like PEP-8. What didn't work: - Nitpicking every detail in a submission, especially style guide rule violations. Students tended to focus on these smallest, concrete mistakes (e.g. spaces between operators) instead of more important conceptual misunderstandings (e.g. when to use recursion). Also poor for confidence. - Falling behind because of perfectionism, which decreased work motivation.