Search, Ranking, RAG

Responsible AI

• Led ML development for Individual Disability Claim Guidance, driving significant advancements in predictive accuracy and operational efficiency.

• Pioneered the company’s first deep dive into model biases and successfully established guidelines and metrics for Responsible AI.

• Successfully introduced two advanced survival modeling frameworks, empowering all product teams with the capability to predict the duration of disability claims accurately.

• Completely revamped the ML training and tuning pipeline, achieving a 2-10x increase in training speed and a marked enhancement in model robustness and reliability.

• Architected and developed Fairness Evaluator, Twitter’s first distributed model performance and fairness evaluation system based on TFX and Apache Beam. Fairness Evaluator has been used throughout the company to evaluate and improve recommendation, ranking, and content moderation models and is a cornerstone of responsible machine learning at Twitter.

• Researched and developed new AI fairness and performance metrics, including the metrics published at ACM FAccT 2022.

• Helped develop IBM’s AI Fairness 360 and AI Explainability 360 toolkits, the first open source ethical AI toolkits in the industry. Helped integrate them into IBM OpenScale. Received IBM outstanding research accomplishment award.

• Researched and developed various prototype systems on active learning, drift detection, and chatbot authoring to inform designs of production systems such as Watson OpenScale, IBM AutoAI, and IBM Watson Assistant. Received IBM outstanding accomplishment award.