Foundational model pretraining research advised by Percy Liang. Built the inference infrastructure for classifying trillions of tokens using 100s of spot TPUs. Answering the question: what is high quality data?

Deep learning systems, LLM and VLM evaluation research (published at MLSys 2024) advised by Ion Stoica and Joey Gonzalez. Worked on serving thousands of LoRA adapters (https://github.com/S-LoRA/S-LoRA) and core member of Chatbot Arena which serves 1M+ requests/month (https://lmsys.org/)

Decreased upscaling latency by 5.1x. Blog: https://www.anyscale.com/blog/autoscale-large-ai-models-faster

Built Beli for travel planning and won a big prize

DGX - Unsupervised anomaly detection of system failuresBuilt a pipeline that extracts real-time software and hardware health metrics from thousands of H100 and A100 clusters and displays the exact component that is down, their firmware version, and the time it occurred. Then, built an unsupervised learning algorithm that predictively detects when systems go down.

Spring 2024 - CS 162: Operating Systems and Systems Programming TAFall 2022 - CS 61A: Structure and Interpretation of Computer Programs TutorSummer 2022 - CS 61A: Structure and Interpretation of Computer Programs TutorSpring 2022 - CS 61A: Structure and Interpretation of Computer Programs Academic Intern

Researching about Understanding Design Thinking and Teaming by Product Managers; advised by Professor Sara Beckman- Performed topic modeling across 600+ survey responses. Used LLMs like BERT for embedding clustering and GPT for cluster summarization. Provided insights on the barriers to implementing design thinking principles in the workplace.- Co-authored paper in submission to California Management Review- Sponsored by DeBenedetti Fund: https://research.berkeley.edu/urap-researchers/christopher-chou/

🏎️ Desk: Built a regularized regression model for futures

Analytics Team: Improving how we track ad impressionsDesigned and developed an endpoint in C++ that parses, validates, and calculates the revenue of over 15M daily impressions for app developers. Implemented asynchronous task execution and write-behind cache to boost end-to-end performance by 40%

Optimizing autonomous driving in unstructured stochastic intersections under Prof. Alexandre Bayen and graduate student Fangyu Wu (Berkeley AI Research, Berkeley DeepDrive)Co-authored paper in submission to International Conference on Robotics Automation 2025

Wrote the investment memo for a pre-seed check in an AI audio survey tool company

User Interactions and Fraud Detection: Helping customers get honest opinions about productsRaised a $4M seed round backed by Vertex Ventures, Sonae IM, and Berkeley Skydeck• Designed a Jinja-based analytics dashboard to display user-interaction data using Firebase, Bootstrap, and Plotly• Developed a fraud detection algorithm through vector-based similarity and unsupervised pattern extraction, tagging 100+ message threads

Landing Page and Conversion Rate Optimization: Connecting international startups with top tier VCsClosed a $60M Fund II• Berkeley SkyDeck is UC Berkeley’s largest startup accelerator, supporting 100+ teams per biannual cohort, each receiving $200K from the Skydeck Fund• Architected the SkyDeck website to facilitate interactions between 1,000+ investors and 16 startups on Demo Day• Conducted a data-driven operation to study user behavior and determine high-impact changes for the website, increasing user engagement by 670% using website heatmaps, funnel analysis, and usability testing• Audited the Canadian Technology Accelerator Demo Day, TTA Global Innovation Showcase, and Skydeck Intern Fair featuring 85+ startups, improving SkyDeck’s deal flow process in a virtual environment• Streamlined onboarding process to manage Skydeck’s growing network of 350 advisors and mentors

Exploring the application of topological semimetals for more efficient electronic devices• Selected as one of 75 students out of 1500 applicants for the Research Mentorship Program• Conducted Research on Effect of Inversion Asymmetry on Quantum Confinement of Dirac Semimetal Cd3As2 with Manik Goyal, a Ph.D. at UCSB • Created a quantum simulation model that manipulated the inversion asymmetry and hybridization of the material using Kwant package in Python• Proposed novel spintronics devices and topological field-effect transistors using Cd3As2• Best Poster Presentation Award at The 9th Global Conference on Materials Science and Engineering (CMSE 2020) and Published in the Journal of Physics: Conference Series