MS Computer Science (Artificial Intelligence & Theoretical Computer Science)

BS Computer Science, BS Chemical Engineering (double degree)

Working on large-scale production ML/data systems with PySpark, Iceberg, Clickhouse, XGBoost/LightGBM, SHAP, MLflow.

Built batch ML prediction framework and feature store with a focus on high performance and fast experimentation, supporting baselines, metrics, reproducibility, explainability, missing/partial features.

Modeling of vehicle sensor and fault code data including transformer model using TensorFlow Official Models NLP library.

Worked on an internal C++ large-scale data processing framework similar to MapReduce/Flume, widely used in Google Ads.

Designed major portions of a framework to listen to Google F1/Spanner changes and write back transformed data, as well as many other features, performance improvements, and regression testing. Have C++ and Python readability.

20% project: Identifying misinformation in social networks using TensorFlow Graph Neural Networks.

Designed and implemented automatic time zone detection for Windows 10, which has since shipped to all users. Inventor on U.S. patent 10,503,124 (Automatic time zone detection in devices).

Hamed Nilforoshan*, Wenli Looi*, Emma Pierson*, Blanca Villanueva, Nic Fishman, Yiling Chen, John Sholar, Beth Redbird, David Grusky, and Jure Leskovec. Human mobility networks reveal increased segregation in large cities. Nature, 2023. [research briefing, project website, code]

Wenli Looi, Manmeet Dhaliwal, Reda Alhajj, and Jon Rokne. Recommender System for Items in Dota 2. IEEE Transactions on Games, 2019. [web-based demo]

On-device ML: Real-time AI camera surveillance on very low-end systems (e.g. Intel GPU). Built on top of raw Vulkan/DirectX shaders for unparalleled performance and stability. Generating OpenCL shaders using TensorFlow Lite and converting to HLSL while optimizing and adding 8-bit quantized weights/activations. Contributing some small improvements back to TensorFlow.

Linux kernel/OpenWrt: Submitted kernel patches to support Qualcomm QCN5502 WiFi SoC based on reverse engineering. Added support for new Wi-Fi routers in open-source firmware, including reverse engineering proprietary image formats.

Competitive programming: ACM ICPC World Finals contestant in 2015 and 2019, and top 500-1000 worldwide in Google Code Jam / Meta Hacker Cup. Gave algorithm lectures and writes contest problems (e.g. on Kattis).

Language learning: Learning several languages. Intercepting Duolingo app traffic with Android TLS certificate bypass and mitmproxy to get unlimited hearts and automatically convert the course content to Anki flashcards.

Home automation: Fully custom system for lights/outlets, furnace, sensors involving PCB design, IP cameras, microcontroller firmware in C (ESP32/ESP8266, Arduino/AVR, STM8/STM32), data storage/analysis/web UI (Elasticsearch, Go, React).

Computer Vision (CS 231N): Quantized GANs for Mobile Image Reconstruction [poster, report]

Deep Generative Models (CS 236): Caption-to-Image Conditional Generative Modeling [poster, report]

Distributed Systems (CS 244B): Google File System From Scratch [report]

Graph ML (CS 224W): Predicting Traffic Congestion on City Road Networks [poster, report]

Machine Learning (CS 229): Analysis of Code Submissions in Programming Contests [poster, report]

Natural Language Processing (CS 224N): Compressed SQuAD 2.0 Model With BERT [report]

Other courses: Convex Optimization (EE 364A), Infomation Retrieval and Web Search (CS 276), Optimization and Algorithmic Paradigms (CS 261), Probabilistic Graphical Models (CS 228), Randomized Algorithms (CS 265)