Rong (Laura) Bao

Model Development & Deployment

• Trained and fine-tuned a LiLT transformer model for multilingual dividends document extraction (~20K docs/month), automating structured information extraction with supervised learning.
• Deployed the model via DSP (Bloomberg internal ML serving platform), reducing data SLA from 10 minutes to 5 seconds, and increasing the STP rate by 58%.
• Built a regression testing framework for dividends ML model to facilitate faster and safer model iterations.

ML Infrastructure & Monitoring

• Designed and built an online metrics system to monitor both model performance and business metrics—such as STP rates and precision/recall/accuracy—for NLP information extraction projects.
• Created a dashboard for real-time metrics visualization and analytics used across five projects in two teams.
• The metrics system replaced manual quality checks, accelerated delivery, and helped teams optimize models to consistently meet STP targets.

Platform & Real-Time Systems

• Collaborated in a cross-team working group as one of three core contributors to architect a scalable and reusable ML onboarding path within Data Technologies, including standardized APIs, data pipeline blueprints, deployment workflows, and monitoring solutions. This paved path successfully enabled a new team to launch their ML project within one month, significantly reducing time to production.
• Developed an entity resolution service using NLP, fuzzy matching, and text similarity to identify securities from company information and descriptions, improving matching accuracy, and reducing manual review efforts.
• Developed a real-time deduplication control service for data pipeline, removing duplicates at the acquisition source level, and preventing downstream duplication. Addressed race conditions to enable consistent data integrity across multiple projects in the team.

Leadership & Community

• Led weekly ML/AI forums across three teams, sharing best practices, project insights, and mentoring team members interested in ML initiatives.
• Mentored a summer intern on Canadian buyback ML extraction project and onboarded three junior developers.
• Actively contributed to recruitment efforts, including technical interviews and hosting university workshops.

• Developed an ML recommendation model for collateral posting management with XGBoost, allowing clients to quickly reconcile disputes with counterparties instead of posting manually. The model was able to achieve above 97% accuracy when back tested on historical data.