mingyang User

I’m a Forward Deployed Engineer focused on deploying AI into real financial workflows. I bridge ambiguous business needs to working systems by clarifying requirements, designing evaluation workflows, and turning failures into product safeguards.

At Cohere, I’m an AI Data Trainer for investment analysis and finance LLM evaluation. I evaluate large-scale LLM outputs using ranking, error analysis, quality auditing, and structured evaluation frameworks—surfacing ambiguity patterns, unsupported claims, and numerical inconsistencies.

Previously at ICBC, I built and maintained Java/Spring Cloud backend services for a large-scale financial platform and improved reliability through architecture, monitoring, and operational debugging. I also developed a Python credit risk modeling system processing 100,000+ applications monthly with SVM, KNN, and Random Forest—improving high-risk detection accuracy by 20% and reducing default rates by 15%.

I engineered Spark, Kafka, and Hadoop pipelines for transaction processing and fraud detection, while integrating ML monitoring and feedback loops to reduce false positives by 30%. I’ve also worked across full-stack AI prototypes—like my Finance Agent Integrity & Reliability Workbench—where evaluation becomes infrastructure, not just a checklist.