Skip to content
Ryder Xue

Ryder Xue

AI products.
Business strategy.

I study business, statistics, and data science at Carnegie Mellon. I turn product questions into working systems, human and LLM evaluations, and commercial decisions.

Explore my work
Ryder Xue in a dark suit and silver tie.
Carnegie Mellon University
Business · Statistics & Data Science

Seeking Summer 2027 internships

Pittsburgh · Shanghai · Hong Kong

Projects

From defining the product problem to choosing what to build and how to evaluate it. These projects connect strategy with working systems.

Frolio

A personal home
for fragrance.

Founder · Product strategy & development · built solo with Claude Code

Feb 2026 — present

I own product strategy and development for a fragrance platform with 150,000 unique fragrances. I connect collection and wear records with three discovery paths: guided taste finding, natural-language search, and daily picks. Each path serves a different decision about what to try next.

Frolio’s collection and discovery website, showing three perfume bottles and the collection introduction.
Guided matching and daily picks use rules. Open-ended scent searches use AI. Collectors can keep notes private, share collections, and review photo suggestions before adding a bottle.
The actual neon pinball playfield, with bumpers, flippers and a launch lane.

Y2K pinball

Python fundamentals · No AI

Built in 15 hours, with no AI assistance. I wrote the original 15-112 game in Python with CMU Graphics to put my programming fundamentals into practice: seven-class object-oriented design, vector math, collision detection, and game-state logic. Play the browser adaptation below.

Play the browser version
Role & context

Independent developer. 15 hours · 15-112 term project.

No AI used. I built the original Python and CMU Graphics game in 15 hours for 15-112, applying Python fundamentals through seven classes and custom physics.

Edge Catcher plans weekly pickups around usable starts, estimated point gain and core roster protection.

Edge Catcher

Product · Fantasy basketball

I built a waiver planner for my ESPN fantasy basketball league. It compares player pickups and drops by estimated weekly point gain, accounts for usable starts and acquisition limits, and protects core players. A weekly scoreboard tracks every team and flags the bottom four. The public demo uses real NBA players with fictional league teams, rosters and scores, and simplifies planning to one streaming roster slot.

Try the demo

Real NBA players, fictional league data. No account or ESPN connection required.

Role & context

Product design & development · built with Codex. October 2026 — present.

Python analysis and a React dashboard combine ESPN league data, legal lineup allocation and a bounded search for weekly streaming plans. The private app refreshes hourly and keeps acquisition advice on hold when source data cannot be verified. The public demo uses real NBA players with fictional league teams, rosters and scores, plus a simplified model with one streaming roster slot.

Research

Questions about basketball, fragrance and the economy, with the methods and limits in view.

Are NBA rookies undervalued in fantasy drafts?

I tested forecasts for all 59 players drafted in 2025. The 20 rookies with verified prices averaged more fantasy points than similarly drafted veterans. Veterans led when I pooled the three earlier classes.

Role & context

I designed the study and built the data pipeline with Codex. The November revision verifies 20 preseason prices and compares rookies with veterans drafted within five positions. Incomplete price coverage limits the conclusion. The paper has not been peer reviewed.

Predicted and actual fantasy points for the 59 rookies, plotted against a line of perfect prediction.

9 of 19

rookies with recorded preseason prices finished above their fantasy draft rank in the original sample.

Does cosmetics spending hold up in a recession?

I compared US cosmetics and clothing spending after nine business-cycle peaks. Cosmetics showed no consistent advantage over the following year.

Role & context

I designed the study and analyzed the spending data with AI assistance. The self-published working paper includes the sensitivity checks and replication materials. The paper has not been peer reviewed.

Nine panels comparing cosmetics and clothing spending around each US business-cycle peak.

4 of 9

US business-cycle peaks after which cosmetics outperformed clothing relative to their earlier trends.

In practice

SQL evaluation pipelines, human and LLM evals, product requirements, and livestream ROI. Analysis that informs what a team does next.

Product / strategy résumé

ByteDance

UGC AI Strategy Intern

May — Aug 2026

I engineered a SQL data-extraction pipeline with a scheduled agent job for recurring evaluation data. I evaluated precision and recall for 50+ human reviewers and four posttrained LLM models across about 20 test sets and their corresponding taxonomy version histories.

More about my work
  • My prompt-engineering and taxonomy revisions contributed to improving the LLM classification workflow’s precision from 71% to 85% and recall from 60% to 85% within one month, using the same comparison test sets with clearer prompts and standards. Posttraining was fixed; the team selected the best-performing model for each revision.
  • I used SQL to diagnose over- and under-recalled content, tracing errors to policy gaps, human-review decisions, and misleading OCR or ASR signals. Those findings guided the next prompt and taxonomy revision, including standards for categories without historical labels.
  • I designed and analyzed a randomized, three-arm survey experiment with 133 participants to measure perceived value at useful-comment densities of 1, 3, or 5 of the top 10. The tested parameters informed human-review recall rules.
  • I proposed a call-to-action content direction, iterated its review standards, and wrote initial PRD drafts for CTA and comment workstreams. Mentors provided input and finalized the requirements.
  • I audited video-category mappings during a search-taxonomy migration from eight levels to three tiers, flagging incompatible and overly broad mappings.

Carnegie Mellon University

Teaching Assistant · 15-112

Aug 2026 — present

I lead weekly recitations, office hours, grading, and AI-assisted programming labs for Fundamentals of Programming and Computer Science (about 15 hours/week), teaching recursion, object-oriented design, and algorithmic problem-solving.

More about my work
  • I lead test-set review for the course’s AI grading workflow, checking cases against grading rubrics. I also develop practice problems and coordinate consistent grading and student feedback.

Nutrition Alligators

Co-Founder & Marketing Manager

Nov 2023 — Nov 2025

I co-founded a nutrition business that generated RMB 1.2M+ in revenue from April 2024 to August 2025. Within a 16-person team, I was responsible for marketing, day-to-day livestream operations, and livestream ROI management. I exited in November 2025.

More about my work
  • I shaped SKU, pricing, and audience strategy, managed and analyzed paid social and 500+ Douyin livestreams, and tracked CPM and ROI to guide marketing decisions. I also participated in 30+ factory and lab visits for product validation and quality review.

Stripe

Product & Operations Intern

Jul — Aug 2024 · Remote

I researched API gateway design, tokenization flows, and payment monetization to understand global payments infrastructure and cybersecurity.

More about my work
  • I worked in a four-person team on competitive and benchmarking analysis of Stripe Radar, presenting our findings to Stripe executives.

A little context

I like finding the connection between a business question and something I can build, measure or test.

Ryder seated on rocks beside a clear natural pool.
Off hours: fragrances, F1, NBA, skiing,
jazz, fashion and poker.

I’m focused on AI product management, strategy, and analytics. My work connects the product problem with how it gets built, evaluated, and operated: SQL pipelines and model evaluations at ByteDance, product decisions at Frolio, and marketing and livestream ROI at Nutrition Alligators.

That range carries into my coursework at Carnegie Mellon, where I’m studying Business Administration with an additional major in Statistics and Data Science.

Carnegie Mellon

B.S. in Business Administration
Additional major in Statistics and Data Science

Class of 2029 · Sophomore · QPA 4.0

Selected coursework
Regression Analysis · Multivariate Analysis · Probability Theory · Finance · Accounting · Programming and Computer Science

Languages
Native English and Mandarin

Ryder Xue

Tools & practice

Analysis
SQL / PostgreSQL · Python (NumPy, pandas, matplotlib) · R / RStudio · Excel
Product & strategy
Product requirements · Discovery flows · SKU and pricing strategy · Livestream ROI
AI evaluation
SQL data pipelines · Human + LLM evals · Prompt engineering · Taxonomy versioning
Building & media
Claude Code · Codex · Cursor · TRAE · Adobe Premiere Pro · After Effects

Beyond the syllabus

Financial Literacy Month

Founder & organizer · Shanghai American School
Apr 2023 — Apr 2025

I designed personal-finance workshops for 80+ elementary students, and worked with the library using Business Club funds to donate 40 books each year.

Shanghai American School Jazz Band

Vice President & Lead Drummer
Aug 2021 — Jun 2025

I led rehearsals and arrangements for 20+ performances a year, mentored about 15 new members annually, and organized jazz jam sessions.

Recognition

  • FBLA China Nationals — 1st in Website Design; top 10 in Entrepreneurship.
  • International Economics Olympiad, nationals — team bronze (top 18%); Cum Laude.
  • 2025 John Locke Institute Essay Competition, Economics — Commendation (top 16%).