01
Structure
Turn ambiguity into an analytical frame
Clarify the business question, identify the real decision, define KPI logic, and separate signal from noise before building anything.
Business Intelligence & AI Systems Designer
I design systems for better business decisions.
I help teams turn ambiguous business problems into structured decision systems through analytics, automation, and AI.
My work sits between business, data, technology, and execution: defining the right question, building KPI and reporting frameworks, standardizing messy workflows, and applying AI where it can become part of real operating systems.
01
Structure, Standardize, Systemize
The common pattern across my work: define the real business question, create shared operating language, then turn repeatable analysis into durable systems.
02
Standardize
Turn fragmented work into shared operating language
Create consistent metrics, reporting logic, data validation, and review processes so teams can compare performance and communicate clearly.
03
Systemize
Turn repeated analysis into durable workflows
Use automation and AI to make reporting, knowledge work, and decision support repeatable instead of one-off analyst effort.
02
Selected Work
Three case patterns that show the same operating model across different contexts: ambiguity to structure, fragmentation to standards, repeated work to systems.
01
Structure
Organization
Artalaxies
Business Analytics Framework for Digital Asset Markets
Multi-source market, user, and operational data integrated into a unified analytics framework for business and executive decision support.
Challenge
Business questions were reactive, KPI definitions lacked consistency, and analysis centered on isolated metrics instead of a repeatable decision framework.
Framework
Market PerformanceUser BehaviorCollection HealthOperational Effectiveness
Before
Reactive metrics, ad-hoc reporting, and questions that stayed at the level of what happened.
After
Standardized KPI logic and repeatable analysis across market, user, collection, and operational dimensions.
Core Contributions
Business problem decomposition
KPI framework and metrics definition
SQL dimensional model and reporting layer
Python ETL and validation workflow
Collection comparison and campaign analysis
Executive decision support
02
Standardize
Organization
Regional analytics · multi-state distribution business
Regional Performance & Target System for Multi-state Distribution
Customer-level order data across a multi-state sales territory, organized into monthly targets and a recurring weekly review that supports regional sales leadership.
Challenge
Sales goals were set as regional totals, and performance was read on mixed bases — order timing against shipment timing, cumulative figures against weekly ones — so the same week could be told several different ways.
Framework
Customer-level TargetsStated Metric BasisWeekly Review CadenceDecline Early Warning
Before
Goals set as regional totals, reports read on mixed bases, and weekly reviews rebuilt from scratch each time.
After
Rep- and account-level monthly targets, one stated metric basis, and a fixed weekly pack and agenda used across regional reviews.
Core Contributions
Monthly rep- and account-level targets cascaded from regional goals
Order-basis and shipment-basis metric definitions
Weekly performance pack template
Customer decline review by ordering cycle
Review agenda from results to follow-up
Analytical support for regional sales leads
03
Systemize
Organization
Piaopiao Travel · Founder, independent AI product
Grounded AI Planning System for International Travel
An AI travel-planning product, built through v1 with a small team and run solo since, where itineraries are assembled from verified flight, hotel, and activity candidates and a maintained country-level visa dataset.
Challenge
Language models write fluent itineraries but invent flights, hotels, and visa rules; a travel plan is only useful when every item and entry requirement in it traces back to checked data.
Framework
Verified InputsVisa Facts from CodeEvaluation Before ReleaseSource Monitoring
Before
Model replies parsed out of free text, and itinerary generation slow enough to time out on mobile.
After
Native tool calling over verified inputs, visa facts supplied by code, releases gated by evals and a human sign-off, and generation fast enough for mobile.
Core Contributions
Decision records ->Tool-calling model restricted to verified candidates
Feature-flagged rollout gated by LLM-judge and adversarial evals
Latency fix splitting fast selection from narrative writing
Weekly audit of official visa sources
Fine-tuned 8B model, evaluated but not deployed
Multi-agent AI coding workflow with a human release gate
03
Thinking
Short operating principles behind how I approach analytics, business intelligence, and AI workflow design.
Metrics are a business language before they are a calculation
A KPI only becomes useful when teams agree what it means, why it matters, who owns it, and which decision it should influence.
Dashboards do not fix unclear decisions
A reporting layer can show movement. A decision system clarifies the question, the threshold, the owner, and the next action.
AI needs workflow, not theater
The strongest AI use cases are practical: organizing knowledge, reducing repetitive work, validating evidence, and helping teams move from messy inputs to useful action.
The bridge is the work
Business, operations, data, and engineering teams often speak different languages. The leverage comes from translating them into one executable system.
04
About
The throughline is business and technology translation: connecting operators, analysts, builders, and decision makers around one executable system.
Background
I work from the business question backward, then build the analytical and technical system around it.
Trained across applied mathematics, data science, public policy, machine learning, and statistics through UC Berkeley and the University of Chicago, with hands-on work in sales analytics, digital-asset market analytics, and building AI products.
Business problem structuring
Decision-oriented analytics
KPI and metrics framework design
Data strategy and BI architecture
Operational intelligence
Workflow automation
AI solution design and integration
Executive communication
Cross-functional translation
05
Decision Diagnostic
A small interactive diagnostic for how an ambiguous business issue should move from question to metric, workflow, and decision.
Decision diagnostic
Choose a messy business problem
The point is not to produce a quick answer. It is to show how the question should move from ambiguity to metrics, workflow, and decision support.
Problem frame
The issue is too broad to analyze directly. It needs to be separated into channel, customer, pricing, conversion, product, and external-market hypotheses.
01
structure
Define the actual decision: diagnose where performance changed and which driver is controllable.
02
standardize
Create a KPI frame across traffic, conversion, retention, order value, product mix, and campaign windows.
03
systemize
Build a recurring review workflow that compares driver movement before leaders ask for another ad-hoc report.
Outcome
Ambiguous revenue discussion becomes driver-based decision support.
06
Contact
For thoughtful conversations around business intelligence, decision systems, operational analytics, or AI workflow design.