Skip to content
View YujunGe07's full-sized avatar

Block or report YujunGe07

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
YujunGe07/README.md
Yujun (Amy) Ge. Stanford CS. AI and ML engineer. Curious across disciplines. Building AI for people.

Email Amy LinkedIn Explore selected work

Stanford CS · Research Assistant at ILIAD Lab / SAIL
Incoming Amazon Software Development Engineer Intern, Summer 2027

I build AI that helps people communicate, learns from complex data, and turns research into useful software. My work spans language and speech, accessibility, computer vision, computational neuroscience, and intelligent agents. I like moving between disciplines and following a question from the first experiment to a working system.

What I bring

Research curiosity across language representations, biological signals, and learning systems. Product thinking through assistive communication and user-facing software. Engineering depth in agents, evaluation, and simulation.

Right now

ILIAD Lab / SAIL   Multimodal learning and generalization across platforms, environments, and tasks.
Common Origin   Founding AI engineer working on world models, simulation interfaces, and repeatable agent experiments.

Previously: dexterous manipulation at AtomBite · browser agents at Kairos UGC · anomaly detection at Haurlan


Selected work

AAC pipeline: listen, suggest, and speak

An AAC prototype combining speech recognition, context-aware text suggestions, and speech output for people with speech or motor impairments.

Whisper DistilGPT2 Flask

Speech and text input, TTS, and interaction history.

Initial BERT compression experiment: 768 to 128 dimensions with 83.4 percent AG News accuracy

Studying how much of a BERT representation we can keep, and whether compression learned on one distribution transfers to another.

Transformers PCA PyTorch

Initial run: 83.4% at 128 dimensions vs. 83.0% at 768. Strict-split and category-shift experiments included.

Molecular wiring study: 17 neuron types and 272 directed pairs

Can gene expression help predict neural connectivity? A public-data pipeline with explicit baselines and held-out tests across Drosophila neuron types.

Neuroscience Ridge regression

17 neuron types. 272 directed pairs. Results and limitations included.

Human pose keypoints mapped to eight humanoid joints

Camera keypoints become constrained robot joint states. YOLO pose estimation, an eight-joint URDF, and ROS 2 / RViz visualization with a desktop preview.

YOLO OpenCV ROS 2

Geometry and URDF contract checks run in CI.

Also on the court: AI-Driven Tennis Coaching uses ViViT embeddings and nearest-neighbor retrieval to compare tennis strokes with professional examples. Includes the research paper and analysis of camera angle, clip duration, and court appearance.

More to explore: ECG signal analysis · Prediction-market matching engine

Research & recognition

Recognition
Regeneron STS Science Talent Search Scholar
MIT URTC 2025 Independent author of an IEEE publication; 4th Place Best Paper Award
IEEE publication First author of research on AI-driven tennis coaching
Tools I build with
Area Toolkit
Learning & perception Python, PyTorch, TensorFlow, Hugging Face, scikit-learn, multimodal models, computer vision
Robotics & simulation MuJoCo, CARLA, Unreal Engine, FlyBody, ManiSkill3, SAPIEN, LeRobot, ROS 2, URDF, FK/IK
Agents & systems Playwright, Docker, tool orchestration, automated evaluation, structured logging, TCP/JSON
Applications & data Java, JavaScript, TypeScript, SQL, Flask, PostgreSQL, Elasticsearch, Git, CI/CD

Different questions. A shared instinct to build.
Helping someone communicate. Understanding a signal. Testing an idea.
I am interested in what AI can make possible across all of them.

yujunge@stanford.edu   /   LinkedIn

Popular repositories Loading

  1. LingsDanceAcademy LingsDanceAcademy Public

    HTML

  2. StarPower StarPower Public

    HTML

  3. TennisAnalysis TennisAnalysis Public

  4. Conflict-Interest Conflict-Interest Public

    Jupyter Notebook

  5. Essay-Brainstorm-Helper Essay-Brainstorm-Helper Public

    JavaScript

  6. V1_AI_ML_AAC V1_AI_ML_AAC Public

    JavaScript