Xiangxi (Sarah) Chen

Chapter 03Projects

Systems, products, and side builds — where the thinking shows up.

Each of these started messy: an ambiguous problem, no obvious shape, and real stakes. They're also the clearest proof that I don't stop at the analysis — I ship the thing.

01

AI Product · WPP Hackathon Winner

Luxe IQ

An AI agent that turns reporting work into client-ready outputs in minutes.

Problem

Strategy and analytics teams were spending too much time manually turning performance data into polished client-facing deliverables. The bottleneck was not insight. It was production speed.

Approach

I built Luxe IQ, an AI agent that generates HTML dashboards and strategy outputs from campaign data and reporting context. The system was designed for practical adoption, not demo value, so the output could move directly into real client workflows.

Outcome

Deployed across 5 to 6 accounts and cut turnaround from weeks to minutes. It also became part of the body of AI work that earned first-place recognition across WPP hackathons.

5–6accounts deployed
Weeks → minutesturnaround time
0 → 1internal AI product
02

Forecasting System · WPP Global Hackathon

CapPilot

A campaign delivery forecasting system that reduced normalized error by 73.5%.

Problem

Media line items regularly under-deliver or over-deliver, eroding campaign performance and client trust. The existing capacity algorithm surfaced problems too late to act on — pacing levers had already expired, and budget was either wasted or left on the table.

Approach

Built CapPilot as a campaign delivery forecasting system using four specialized XGBoost models split by DSP and buy type. The feature design focused on levers planners can actually act on, so the forecast could improve optimization instead of just describing risk.

Outcome

Reduced normalized forecasting error by 73.5%, giving teams earlier risk detection and more reliable spend decisions. The work was recognized with a first-place finish in the WPP Global Hackathon in the U.S.

-73.5%forecasting error
4specialized XGBoost models
US #1WPP Global Hackathon
03

Analytics Leadership · WPP Media

Measurement Systems That Move Budgets

Measurement, migration, MMM automation, and lift studies built to influence real spend decisions.

Problem

Cross-market campaigns generated noisy performance data, scattered across platforms on lagging refresh cadences. Leadership was making budget calls on what was visible, not on what was true.

Approach

Built analytics foundations for new accounts, led a 90-day Datorama-to-Power BI migration, defined KPI and benchmark frameworks across 10+ brands, and automated MMM data collection with reusable SQL views for Circana and Nielsen workflows.

Outcome

Created a measurement layer that moved faster and reached farther: automated processes that dropped from about a week to under 20 minutes, lift studies that informed quarterly recommendations, and systems leadership teams could actually trust.

10+brands and global campaigns
<20 minMMM prep time
6–8 / qtrlift studies run
04

Side Project · Product + Brand

An internet-native resale concept, built like a product story.

Problem

Most resale products treat impulse buys as commodity inventory. The emotional layer — the absurdity, the regret, the story — goes unused and the experience feels anonymous.

Approach

Defined the product tension, built the brand world, and shaped the interaction design to feel like a piece of internet culture rather than a marketplace. Solo design and build.

Outcome

A working concept that shows how I think beyond analysis alone: define the user tension, sharpen the narrative, and build a system people instantly understand.

Solodesign + build
Brand-ledproduct narrative
Sidepersonal R&D
05

Side Project · Career Tool

VizUp

An app that helps workplace professionals grow their visibility.

Problem

Workplace visibility — being known for the right work — is usually treated as luck or social skill. Most people who struggle with it aren't short on competence; they're short on feedback, structure, and a way to practice.

Approach

Designed and built VizUp as an opinionated tool for that gap: prompts, practices, check-ins, and AI-driven coaching built around real workplace dynamics. Solo design and build, end-to-end.

Outcome

Now live on the App Store. It shows the pattern I care about most: find an underserved behavioral problem, design around it, and ship it end-to-end without waiting for a mandate.

Solodesign + build
AI-drivenpersonalized coaching
App Storelaunched

Want to talk about a project, a role, or a weird idea?