Inference optimization

SFT & RL service

TL of Knowledge Graph powered GenAI Trust Safety.

• Built Trust Risk Engine with low latency configurable C++ libraries which works for both online features and batch jobs.

• Provide precomputed trust signals and trust policies for multi-surface GenAI products (Search AI Overview, Gemini, Cloud VertexAI, etc)

• Improved qualities to achieve 99%+ recall with decent precision for 2024 Election protections.

• Built LLM based Search Quality and KG classifiers.

PAST

• Knowledge Graph Quality and Trust: improve data indexing trust evaluation and human review platform.

• Knowledge Graph coverage and quality improvement with user feedback

• Knowledge Graph I18n via smart language fallback and machine translations.

Software engineer intern in Wepay's Risk engineer team. Worked on Fraud Model Training pipeline migration.

• Refactored ETL and model training pipeline from local Python script to Airflow;

• Enabled data IO on BigQuery and optimized process with parallelism.

The migration reduced daily model training process time by 70%, from 9-13 hours to 3-4 hours. It eliminates the performance bottlenecks and single point of failure in the old approach.

Shipping containers optimization based on LP, network flow and different heuristic search strategies.

My work: Data insight. Computed historic stock and cost based on different logic using Python, HANA SQLscripts and L language. These are KPIs in new version of the product and parameters for resource planning optimization.