Frank Fu, data-centric product dreamer. Goal oriented, total clarity, real adoption.
— DATA-CENTRIC PRODUCT DREAMER
01 — ABOUTLOS ANGELES, CA
I build the thing the analysis points to.
Finding the cause is half the job. The other half is deciding what's worth fixing and shipping something people will actually use: a model01, a dashboard02, an agent03, an app04.
The best ones stop being tools and become how the work runs.
SOURCES
- [1] MASTER OF BUSINESS ANALYTICS — UCLA ANDERSON SCHOOL OF MANAGEMENT2025.09 — 2026.12
- [2] APPLIED STATISTICS & UX DESIGN — UNIVERSITY OF TORONTO2021.09 — 2025.05
Curiosity. Learn. On repeat.
02 — EXPERTISEQUERY ME
-- what I work with, and the tools behind it SELECT capability, tools, domain FROM frank.expertise;
capabilitytoolsdomain
0Problem solvingCoffee / Ask why ×5 / Jot it down / Solve the real oneALL
1Data analyticsSQL / Python / R / ExcelDATA
2Data engineeringDatabricks / Snowflake / Fabric / PySpark / Medallion ETL / Semantic modelingDATA
3Machine learningscikit-learn / Random forest / XGBoost / Feature engineering / Cross-validation / ClusteringDATA
4OptimizationGurobi / Integer programming / ForecastingDATA
5ExperimentationA/B testing / Causal inference / RDD / GA4DATA
6VisualizationPower BI / DAX / TableauDATA
7AI engineeringLangGraph / LLM APIs / RAG / MCP / Evals / AI governanceAI
8ProductPRDs / User flows / Figma / AgilePRODUCT
9Frontend/backendReact / Next.js / Azure / APIsPRODUCT
10AutomationPower Automate / Power Query / Databricks GeniePRODUCT
[11 rows x 3 columns]
PROJECTSLIVE — see live progress on every project
The scorecard that builds itself.
SAP data piped bronze to gold and refreshed on schedule. An agent team runs the root cause analysis and writes the actions into the scorecard. Build time cut 85%.One customer, 360°.
CRM, orders and fulfilment joined into one customer. Demand is forecast, segments act on their own, and agents escalate the exceptions. Repeat orders up ~11%.A Lean line you can run.
A spreadsheet-run production exercise rebuilt as a React app on Azure with a live what-if simulator.Open the riskiest trial first.
Regulatory filings unified Spark to Snowflake, then every trial ranked by termination risk.Throw anything in. Never forget it.
Anything I throw in stays findable and linked, so I can draw connections between ideas. Ask any time; every answer cites its source.More on the way.
Two builds in progress. This slot fills as they ship.
04 — APPROACHFIVE RULES
- 01WHAT, WHY NOW, SO WHATThree questions before I open anything. If I can't answer why now and so what, I'm not ready to start.
- 02FIX THE CAUSEPatching the symptom just makes a workflow heavier. I trace the problem to where it starts and fix it there, even when that means rebuilding the thing.
- 03A NUMBER ISN'T AN ANSWERA status figure says what happened. I don't hand it over until it also says why, and what to do about it.
- 04ADOPTION IS THE PROOFDelivery isn't the finish line. I'd rather measure whether people came back to the thing than whether I shipped it on time.
- 05SHIP, LEARN, ITERATEShipping is how I learn what I built. Version one is a hypothesis. I rebuild it on what people do with it, not what they say about it.
05 — STUDIES
WHAT I TOOK APART
Coursework and independent analysis.
