Built the team and the roadmap
Grew data science and analytics from zero to fifteen. I run hiring, coaching, performance reviews and compensation, and own a multi-year roadmap agreed with executives, Product and Engineering.
Current role
Beatdapp
- Head of Data ScienceAug 2023 to now
- Data ScientistOct 2021 to Aug 2023
Beatdapp builds fraud detection and analytics for the music streaming industry. I joined as a data scientist, built the core fraud-detection system, and now lead the data science function I grew from zero to fifteen people.
80%+
less streaming fraud across multiple platforms, from the anti-fraud initiatives I led
0 → 15
data scientists and analysts hired, coached and led
Millions
saved for clients in falsely claimed royalties
Fraud-detection partners include
- Universal Music
- iHeartRadio
- The MLC
- SoundExchange
- Audiomack
Plus other partners whose names can't be shared publicly.
Fraud detection in production
Built the core detection system in Python, SQL and Google Cloud. It catches botnets and account takeovers while keeping false positives low.
Human in the loop
Designed the active-learning pipeline: annotation guidelines, label-quality checks and adjudication of disputed fraud labels, now extended to AI agents.
AI music and AI agents
Lead the data scientists and ML engineers building AI-generated music detection, plus an internal program that puts AI agents on streaming and royalty-claimant fraud.
Evaluation and governance
Controlled tests, holdout validation and precision/recall trade-offs before release; KPIs, monitoring thresholds and escalation protocols after it.
Clients and executives
Run recurring reviews with executives and major rights holders, and turn the analysis into product priorities. Self-service KPI dashboards in Power BI, Tableau, Looker and Redash keep the numbers in everyday decisions.
Before Beatdapp
Memory research at WashU, public service in Taiwan, then dementia cohort data in Calgary.
-
2020 to 2021
Data Analyst
Hotchkiss Brain Institute, University of Calgary
Statistical and machine-learning analysis of large dementia cohorts in R and Python, across three to four concurrent projects and seven papers. Mentored graduate and undergraduate students in analysis and visualization.
-
2018 to 2019
Supervisor, civil service
National Immigration Agency, Taiwan Taoyuan International Airport
Supervised and trained 40 servicemen in a high-volume airport. Automated passenger-volume reporting and administrative workflows in Python and Excel, cutting the workload in half.
-
2012 to 2018
PhD researcher
Washington University in St. Louis
Designed behavioral and fMRI experiments on human memory, collected MRI data at the School of Medicine, and built reproducible pipelines in Python, R and MATLAB with mixed-effects models and causal inference.
Education
- PhD, Psychological and Brain SciencesWashington University in St. Louis, 2018. Cognitive, Computational and Systems Neuroscience pathway.
- B.S., Psychology and BiologyUniversity of North Carolina at Chapel Hill, 2012
Toolkit
- ModelingPython, PyTorch, TensorFlow, scikit-learn, R, experimental design, active learning
- DataSQL, BigQuery, Google Cloud, AWS, ETL
- ReportingPower BI, Tableau, Looker, Redash
- DeliveryJira, GitHub, Claude Code, Codex, Cursor
Research
My PhD used fMRI to study how people retrieve memories. At the University of Calgary I worked on behavioral and blood markers of early Alzheimer's disease.
Are There Multiple Kinds of Episodic Memory? An fMRI Investigation Comparing Autobiographical and Recognition Memory Tasks
700+citations
10h-index
Dementia and mild behavioral impairment
- Plasma phosphorylated tau at threonine 181 and neuropsychiatric symptoms in preclinical and prodromal Alzheimer diseaseGhahremani M, Wang M, Chen HY, Zetterberg H, Smith E, Ismail Z. Neurology, 2023
- Vitamin D supplementation and incident dementia: Effects of sex, APOE, and baseline cognitive statusGhahremani M, Smith EE, Chen HY, Creese B, Goodarzi Z, Ismail Z. Alzheimer's & Dementia: DADM, 2023
- Cognitive profile of people with mild behavioral impairment in Brain Health Registry participantsKassam F, Chen H, Nosheny RL, McGirr A, et al. International Psychogeriatrics, 2023
- Plasma β-amyloid in mild behavioural impairment: neuropsychiatric symptoms on the Alzheimer's continuumMiao R, Chen HY, Gill S, Naude J, Smith EE, Ismail Z. Journal of Geriatric Psychiatry and Neurology, 2022
- Dual-task gait and mild behavioral impairment: The interface between non-cognitive dementia markersGuan DX, Chen HY, Camicioli R, Montero-Odasso M, Smith EE, Ismail Z. Experimental Gerontology, 2022
- The relationship between hearing and mild behavioral impairment and the influence of sexGosselin P, Guan DX, Chen HY, Pichora-Fuller MK, et al. Journal of Alzheimer's Disease Reports, 2022
- White matter hyperintensities and mild behavioral impairment: Findings from the MEMENTO cohort studyMiao R, Chen HY, Robert P, Smith EE, Ismail Z. Cerebral Circulation: Cognition and Behavior, 2021
Memory and fMRI
- Task-related and resting-state fMRI identify distinct networks that preferentially support remembering the past and imagining the futureGilmore AW, Nelson SM, Chen HY, McDermott KB. Neuropsychologia, 2018
Projects
What I build outside work, mostly with AI coding agents as the team.
生逢三國Born Among Kingdoms
A Three Kingdoms historical simulation I am building in Godot, focused on character relationships and player-driven stories.
Code owns every rule, number and outcome. Local language models only voice the characters and propose changes, which code validates before anything happens.
- The battle prototype runs signature moves, sieges and cavalry on one simulation core, with figures, poses and ink shaders built in code.
- Built with Claude Code, Codex and Cursor working in separate implementation and review lanes, with pre-merge checks.
Molody
AI background music inspired by molecules.
Built with Chia-Hsiu (Justin) Chen. Each track is generated from a molecular structure, and listeners browse by genre or by chemistry, from the periodic table to proteins.
- Mixtures play as bands: each ingredient takes a role, lead, bass or inner voice, from its chemistry.
- AI generation with ACE-Step, alongside my own algorithmic music generation in PyTorch and TensorFlow.