Tech lead for phishing detection. Responsible for ideation, proof-of-concept demonstration, and production deployment of novel applications of machine learning for detecting phishing webpages.• Developed an AI-based prefiltering pipeline for detecting deepfake scam videos in customer traffic (patent pending). Currently detecting dozens of deepfake scam videos per week in production.• Led a cross-team adversarial AI project using generative AI to create novel variants of malicious JavaScript that evade detection. Used the resulting samples to re-train a malicious JavaScript classifier, resulting in a 10% boost in real-world malware detection rate. (Patent pending, showcased at Black Hat 2024)• Developed a lightweight deep learning model to detect phishing webpages based on the HTML content of the webpage (patent pending). Industry-first use of deep learning for inline detection of patient-zero phishing webpages. Currently detecting ~100K phishing webpages per week in production• Designed and deployed a URL-based prefiltering algorithm, OCR pipeline, and multimodal deep learning model for detecting highly evasive phishing webpages hosted on legitimate SaaS platforms (patent pending)• Primary inventor for 5+ patent applications related to novel applications of AI for phishing/scam detection• Gave company-wide tech talks/training sessions regarding AI/ML for phishing detection• Produced research on global phishing trends that was picked up in major global news outlets, including the Washington Post and POLITICO

Featured on PBS, the Bill & Melinda Gates Foundation, OpenIDEO, and the Westly Foundation, Duet is a tech nonprofit that enables donors to provide meaningful aid to vulnerable populations in a more dignified, efficient, and personal way: https://giveduet.org

Publications:• L Hu, C Robinson, and B Dilkina. 2021. Model Generalization in Deep Learning Applications for Land Cover Mapping. In Fragile Earth ’21: KDD 2021 Workshop. https://arxiv.org/abs/2008.10351Won 1st place in the 2020 IEEE GRSS Land-Cover competition (http://www.grss-ieee.org/community/technical-committees/data-fusion/data-fusion-contest/).Published a codebase for link prediction within social networks. Over 200 stars on GitHub (https://github.com/lucashu1/link-prediction).Labeled UAV footage from wildlife parks to train R-CNN models to detect poachers/endangered animals (https://www.cais.usc.edu/projects/ai-for-conservation/).

Implemented a production-ready deep learning model that detects and categorizes sensitive and/or confidential files (e.g. company source code and financial records). Achieved over 99% accuracy on a test set of 5,000 documents.

USC's leading AI/Machine Learning student organization — dedicated to furthering AI for social good.Oversaw pro-bono AI/ML project teams: worked with clients including the LA Mayor's Office, UCSD Kawasaki Disease Research Center, etc.Designed an semester-long introductory deep learning curriculum for new members.http://caisplusplus.usc.edu

Built a semi-supervised convolutional autoencoder for image feature extraction, allowing images to be incorporated into Falkonry’s event-detection pipeline; reduced the classification error rate by over 75% compared to previous image featurization techniques.Developed an internal Python API and ready-to-use Jupyter notebooks for performing Exploratory Data Analysis on multivariate time-series data, enabling rapid data experimentation/analysis from within the live app environment.

Analyzed web-crawler data to create a recommender system for suggesting content delivery network (CDN) services to various websites (as part of USC Viterbi’s Research Overseas Program) (https://github.com/lucashu1/CDN-RecSys)

Collaborated with IT consultant to implement various informatics technologies (e.g. Pipeline Pilot, SQL Developer); Wrote web service protocols in Pipeline Pilot to enable automatic prediction of compounds’ chemical properties