Amber

Completed both degrees in two years by doubling the course load while running two research streams in parallel. Data Science gave me the tools; Economics gave me the questions worth pointing them at. Published first-author on adverse drug reaction prediction through URAP before graduating — the proof that execution and depth don't have to trade off.

Published Wanyu Zhu et al., 2023, Improvements in Adverse Drug Reaction Prediction (Journal of Physics: Conference Series). Built a multi-threaded Scrapy pipeline that extracted 1M+ comments across 500+ college-related topics on Zhihu, cutting extraction time 50% via async processing. On the ADR work, tuned Random Forest, Gradient Boosting, and SVM with SHAP-based interpretability. Peer review taught me more about ML practice than any coursework.

Four-month program developing the muscles to reason from competitive structure and regulation to price. Wrote and presented a full industry thesis, distilled from primary research into actionable recommendations pitched to the firm's investment committee. This is where the quant-thinker mindset actually clicked — without the language of finance, the technical side wouldn't have a target to point at.

Four months deep in cluster ops: maintained and deployed gRPC, DL, and HPC applications on Kubernetes clusters. When HDFS I/O bottlenecked on DL workloads with many small files, I wrote 1,735 lines of C using POSIX truncate/wrap primitives, NFS mount, and loop devices — a middleware that wrapped small files without breaking POSIX semantics or consistency. Used Perf, Netperf, mdtest, and ioprof to trace the real source: centralized metadata management in the HDFS backend.

Owned the ingest-to-feature pipeline. Combined Quantumera's existing dataset with real-time public-transit APIs into S3; used Redis via ElasticCache to accelerate hot paths, DynamoDB for business logic, AWS Step Functions + EMR for periodic batch cleaning. Features landed in ElasticSearch for downstream ML. Deployed on ECS with auto-scaling and wrote unit, integration, and load tests to validate under real traffic. This is where I learned what production AI infrastructure actually looks like — the glamour is in the dashboards, the work is in the plumbing.

One-year master's focused on the math under every system I've built: stochastic processes, convex optimization, decision theory. The applied capstone pairs students with a real industry problem — supply-chain optimization in my case. This stage is about welding rigor to shipping — the two modes I'd kept on separate tracks until now.