Derek Cheng

Leads technical direction for production machine learning and generative AI on Stack Overflow's developer knowledge platform, setting architecture standards, evaluation practices, and launch criteria. Owns the hybrid ML + generative AI architecture behind Question Assistant, including TF-IDF classifiers and Google Gemini feedback generation.

Co-developed Question Assistant, an AI feature that reviews draft questions and helps developers add missing context, attempted solutions, error details, and minimal reproducible examples. Ran A/B tests measuring ~12% higher question success rates, leading to the feature's launch to all question askers.

Built computer vision and NLP models that auto-populate item metadata from listing photos and text, cutting average listing time 35% and raising production listing volume 20%. Designed and built an item recommendation service on Google Vertex AI using vector similarity search, sustaining 1,000+ requests per second at under 50ms latency.

Developed GPU-accelerated Automatic Target Recognition for a defense air-vehicle platform using residual CNNs in TensorFlow and Python, detecting and classifying targets in Synthetic Aperture Radar imagery. Developed supervised and unsupervised maritime anomaly detection on large, imbalanced vessel-tracking data, extracting time-series features with TSFRESH.

Built a modeling and simulation framework to evaluate Sense and Avoid for a defense Unmanned Aerial System, with C++ interfaces wrapping ACAS-Xu logic and connecting to simulated flight controller, ADS-B, TCAS, and radar sensors. Created visualization tools for UAS behavior and collision-avoidance encounters.

Developed an exportable C# flight dynamics model based on six-degrees-of-freedom performance tables. Built MATLAB automation to plot flight-log data and identify landing-gear and braking faults, replacing manual review of individual logs.

Pursued a Master of Science in Data Analytics with a focus on Computational Data Analytics.