Computer Science @ UC Berkeley · AI Researcher · Software Engineer

"What is the most important step a man can take? It's not the first one, is it? It's the next one. Always the next step."
NeurIPS2024 acceptance & presentation
IEEE ICMI2026 first author
1stNJX & 1435 Capital Hackathon
YCStartup School '26

01experience

Software Engineer

Dandelion Health Inc.

Jun 2026 – Aug 2026 Remote
  • Built the AHA AI Lab Assessment Platform with the American Heart Association: interactive tools for exploring model performance across real-world patient cohorts and hospital populations.
  • Worked on the Go + Kubernetes backend that connects the web portal to multimodal patient data: EHR, clinical notes, medical imaging, and ECG waveforms.
  • Added fingerprint-based stale-data safeguards, TanStack Query data loading, and Vitest/Playwright integration and e2e tests.
GoKubernetesTypeScriptTanStack QueryPlaywright

AI Researcher

University of California, Berkeley

Jun 2024 – Present Remote
  • Advised by Berkeley researcher Jonathan Lu.
  • Tested multilingual LLMs on claim verification across 15 languages. They underperformed on low-resource languages, which are underrepresented in research and training data.
  • Proposed a new Translation Bias metric, built on COMETKIWI, that scores accuracy and translation quality across languages.
LLMsMultilingual NLPEvaluation

Research Scholar

Carnegie Mellon University

Jun 2025 – Aug 2025 Pittsburgh, PA
  • Built a Q&A chatbot on Ollama that answers from personalized datasets using RAG.
  • Worked with Professor Neil Simonetti on an Ultimate Tic-Tac-Toe AI (minimax, alpha-beta pruning, transposition table, parallelized on the Bridges-2 supercomputer) that beats the hardest difficulty of a widely used engine.
RAGOllamaHPCGame AI

02research & publications

IEEE ICMI 2026

First-author paper on multilingual claim verification and translation bias in LLMs.

NeurIPS 2024

2nd Workshop on Attributing Model Behavior at Scale: accepted and presented.

NeurIPS

Safe Generative AI Workshop: accepted.

SoCalNLP

SoCalNLP Symposium: accepted.

03projects

1st Place

GuardLine

NJX & 1435 Capital Hackathon

A real-time phone-screening system that routes Twilio calls through an OpenAI Realtime receptionist and a GPT-4o-mini risk classifier, which approves, blocks, or holds AI-driven scam calls.

Also has a whitelist, multi-round questioning, push alerts, SQLite call logging, and a live security dashboard.

ReactNode.jsTypeScriptTwilio

Ultimate Tic-Tac-Toe AI

Engine + browser game

Minimax with alpha-beta pruning and transposition-table caching. It beats the highest difficulty of Michael Xing's widely used open-source engine.

Search runs in parallel on NSF's Bridges-2 supercomputer, scaling from 1 to 256 threads per RM node.

JavaJavaScriptMultithreading
play it in your browser →

04honors & achievements

The Leadership Award$11,600
Pfizer OSA Scholar$12,000
PA DECA Scholarship$1,000
1600 SATPerfect score
3× AIME QualifierMAA
USACO GoldCompetitive programming
USABO SemifinalistTop 225 · Honorable Mention
YC Startup School2026
VEX Robotics3× Eastern PA skills champion · Worlds qualifier

05education & skills

University of California, Berkeley

B.A. Computer Science · College of Computing, Data Science, and Society

Coursework

Discrete Math & ProbabilitySICP (CS 61A)Foundations of Data ScienceMultivariable CalculusLinear AlgebraLogic

Organizations

Alumni Scholars Program

Before Berkeley

  • Pennsylvania Governor's School for the Sciences @ CMU
  • Research intern @ Temple University, advised by Prof. Stephen MacNeil
  • VEX Robotics Team 6121C programmer · varsity squash

Technical Skills

Languages

PythonJavaTypeScript/JavaScriptC/C++GoSQLHTML/CSSRMATLAB

Frameworks

FastAPIFlaskNext.jsReactNode.jsVitePandasNumPyscikit-learn

Infra & Tools

DockerKubernetesAWSAzureSnowflakeLinuxGitCUDA

Databases

RedisDynamoDBSQLite

06chat with me

my venmo is @ryanfding btw 😛