About
Bio
Seattle, WAI'm a machine learning engineer who builds production ML and agentic AI systems — the kind that price real assets, answer real customers, and page someone when they break. For the last few years that has meant owning pricing end to end at Alt: models, infrastructure, tooling, and the agents that sit on top.
I started in data science across banks and insurers — HSBC's R&D lab, RBC, CI, Sun Life — doing churn models, credit risk, synthetic data, and experiment design. What pulled me toward engineering was watching good models die in handoff: a model is not a prediction so much as a promise someone has to keep, and keeping it means owning the system around it.
Outside work I mentor career-changers into analytics and build small systems for myself — eval harnesses, monitors, agents — because the fastest way to understand a tool is to be its only user.
Experience
Since 2019- 2022 — Present
Senior Machine Learning Engineer · Alt
- Built a user-facing agentic selling product — comparable research, LLM-judge selection, and a self-correcting generator producing a per-quote sell pitch. First phase more than doubled offer acceptance (11.5% → 26%).
- Engineered the model-evaluation triage that picks the best pricing model per asset bucket — raised the auto-priced rate by more than 20%. Patent pending, named inventor.
- Architected the event-driven backbone (SQS + Lambda) that triggers the agentic selling and triage systems.
- Optimized time-series-cache infrastructure — cut monthly compute costs 43% ($95K/year) while improving training and serving reliability.
- Rebuilt the expert-pricing tool on a secure, auditable architecture — 2× processing speed, ~4× quote throughput at rollout.
- Sole engineering owner of pricing for 9+ months across four product surfaces — roadmap, architecture, on-call, and stakeholder alignment as a one-person team.
- 2020 — 2022
Full-Stack Data Scientist · HSBC, R&D Lab
- Prototyped a GAN conditional data-generation model for transactions data under regulatory and privacy constraints; automated ingestion of synthetic network data into Neo4j for all teams.
- Deployed a Python + Scala graph builder and EDA tool used across all projects — cut ETL and exploration time on payments data by 25%.
- Led the Canada Innovations branch-analytics project — surfaced 200+ inside-sales opportunities and optimized branch-level client management.
- 2020
Data Scientist · RBC
- Modeled customer churn at the transaction level with Cox survival analysis — risk scores and urgency ranks advisors used to retain high-risk clients proactively.
- 2019
Junior Data Scientist · CI Global Asset Management
- Diagnosed and resolved data leakage, class imbalance, and overfitting in purchase/redemption prediction — improved model validity.
- 2019
Data Science Analyst · Sun Life
- Designed an A/B testing framework with non-compliance corrections to protect experiment validity ahead of full-scale deployment.
- 2023 — 2026
Mentor — Data Analytics · CareerFoundry
- Mentored students through the Data Analytics Career Transition Program — project reviews, mock interviews, and coaching on data-driven thinking.
Education
- 2018 — 2020Master of Economics · University of Waterloo
- 2014 — 2018Bachelors of Mathematical Economics · University of Waterloo
Toolbox
- ModelingClassification · Regression · Deep learning (PyTorch) · Survival & uplift · Causal inference · Interpretability (SHAP, LIME)
- LanguagesPython · SQL · Spark (PySpark, Scala) · JavaScript
- AI systemsLLM agents · Agentic systems · Orchestration · Event-driven architecture (SQS, Lambda) · Hybrid ML + LLM pipelines
- Eval & qualityLLM-as-judge harnesses · Langfuse · MLflow · A/B testing · Monitoring & drift detection
- Data & infraSnowflake · Airflow · Postgres · Neo4j · AWS · Docker · Terraform · CI/CD · Datadog
Always happy to talk ML, agents, and systems.
sophie.fc.yin@gmail.com