RUIXIN (LUCKIN) YANG
MS in Marketing Science, Columbia Business School. Graduating Dec 2026.
Not one industry, but a map
I haven't worked in a single vertical. I've worked across marketplace, software, digital, retail, mobility, travel, delivery, gaming and beauty, with the heaviest, most technical experience concentrated in technology.
Product. Growth Marketing. Analytics.
Three pillars, one working style. Pick a pillar, then click a case to see the work behind it.
AI → Workflow → Build → Thinking
Not "I use ChatGPT", but how AI changes the way I work, what I've built with it, and how I reason about it.
NYC Live Data
From live operational data to daily decision monitoring. Connected our data environment with Claude to identify success metrics, pull the relevant SQL, visualize live NYC operational data, and create a daily Slack monitoring workflow.
▾Connect
Redash + our Claude account.
Identify
Identify the business success metrics that need to be monitored. Use Claude to help pull out and structure the relevant SQL queries.
Query
Run the relevant SQL against the live NYC operational data.
Visualize
Turn query results into a dashboard / visual monitoring layer: KPI cards, trend charts, comparison views, operational metrics.
Monitor
Set up a daily Slack monitoring workflow. Important changes in the data surface automatically, rather than requiring someone to repeatedly check the dashboard.
Human
- Business problem
- Metric definition
- Workflow design
- Decision logic
AI
- Query assistance
- SQL extraction
- Workflow acceleration
- Data interpretation
System
- Dashboard
- Visualization
- Slack monitoring
- Daily reporting
Competitor Pricing Tracking
▾From manual Excel checks to a collaborative pricing tool. Used vibe coding to replace manual Uber route price checks with a tool that centralizes routes, automates data capture, validates abnormal prices, tracks team completion, and compares Uber pricing with TADA.
We needed to develop TADA's pricing model, which meant continuously checking Uber prices across major NYC routes. Originally everyone on the team manually opened Uber, searched a route, checked the price, and copied it into Excel, which was time-consuming, easy to forget, and hard to consolidate across the team.
Saved Routes
The tool stores the routes that need to be checked. Instead of manually searching for every route, one click opens Uber directly, cutting navigation and search friction.
One-Click Save
Checking a price and saving it now auto-updates the shared CSV. No one needs to manually upload or consolidate a spreadsheet anymore.
Team Completion
The tool tracks each person's daily completion, creating lightweight accountability: visibility that drives a higher completion rate.
Price Range Validation
Each route has an expected price range. An entry outside that range is flagged, catching abnormal prices, accidental input, and data quality issues.
TADA vs. Uber
After a route is submitted, the tool automatically compares the Uber price against TADA's: collect, validate, compare.
| JFK → Midtown | $34.20 |
| LGA → Downtown | $28.50 |
| Newark → Times Sq | $41.00 |
Theory
MS coursework covering deep learning, CNNs, RNNs/LSTMs, reinforcement learning, and Transformers, alongside a Digital Marketplaces course connecting surge pricing theory, LP dispatch optimization, and backpressure dynamic pricing.
Understanding → Application
Are All Prediction Markets Equally Forecastable?
MS in AI research paper: segmented deep learning study of 171 resolved Polymarket contracts. Applied GAF encoding + pretrained ResNet-50/AlexNet, with a 1D-CNN ablation isolating the value of image encoding vs. raw sequences. Found unconditional metrics are confounded by price staleness (~61% zero returns); under a conditional evaluation framework, sports contracts reach r = 0.668 near resolution while crypto collapses to r = 0.030 outside the dense mid-price region.
