WGWYATT GREENE

I WAS JUSTCURIOUS

The letters contain fragments of source code, system plans, Roots in Sign, and the home AI lab.

01 / WHO I AM

Wyatt Greene smiling

I'm Wyatt Greene. I build AI systems, machine-learning infrastructure, and developer tools.

Depending on the day, I might be working on production systems at IBM, testing generated quantum code, working on HRM reasoning models, or fixing something in the server rack at home. I tend to follow a problem as far down the stack as it goes.

CURRENT ROLEIBM UNIVERSITY CO-OP

Cognitive Tooling

FOUNDERGRAYAREA LABS

AI systems, developer tools, and research.

EDUCATIONMSU DENVER

Double-majoring in Data Science & Machine Learning and Applied Mathematics.

CURRENT RESEARCHHRM REASONING MODELS

Creating HRM reasoning models.

GRAYGATE / TERMINAL DEMOVERIFIED QISKIT.
NOT GUESSES.

Watch one request become executable Qiskit, run inside Aer, and return only after the measured result matches the intended circuit.

GRAYGATEQISKIT / AER
PLAIN LANGUAGE IN→EXECUTED QISKIT OUT

GrayGate types a request for a Bell-state circuit, generates Qiskit, runs it in Aer, confirms that the Bell-state correlation matched, and saves the verified program.

QUANTUM CODE GENERATION / EXECUTION / REPAIR

02.1 / SELECTED PROJECT

GRAYGATE — Generate Qiskit. Run it. Repair what fails.

GrayGate takes a plain-language quantum programming request and returns Qiskit code that it has executed and checked. If execution fails, the failure is fed back into a targeted repair pass before the program is returned.

VIEW THE LIVE PRODUCT ↗
ROLEFOUNDER / BUILDER
PRODUCTQUANTUM CODE GENERATION
EVALUATION141 / 151 RE-GRADED RUN
TRAINING LOOP / FORWARD + BACKWARDTRAIN IN PYTHON.
RUN IN C.

A complete machine-learning framework with tensors, autograd, layers, losses, optimizers, data utilities, and classical ML.

Sequential GreyML forward and backward passA fully connected model with eight inputs, four hidden layers, and five outputs. Signals move one layer at a time to the outputs before gradients return through the graph.INPUT / 8HIDDEN 01 / 12HIDDEN 02 / 12HIDDEN 03 / 12HIDDEN 04 / 8OUTPUT / 5
01PYTHON APINUMPY-FRIENDLY MODEL AUTHORING
02C17 TENSOR COREN-D ARRAYS / MATH KERNELS
03AUTOGRADDYNAMIC COMPUTATION GRAPHS
04LAYERSTRAINABLE NEURAL MODULES
05LOSSESOBJECTIVES FOR TRAINING
06OPTIMIZERSPARAMETER UPDATE RULES
07DATA UTILITIESBATCHING / PREPROCESSING
08CLASSICAL MLALGORITHMS BEYOND NNS

02.2 / SELECTED PROJECT

GREYML — A complete training stack with a C17 core.

GreyML is a trainable machine-learning framework with a C17 computational core and a NumPy-friendly Python API. It includes tensors, math kernels, autograd, neural-network layers, losses, optimizers, data utilities, and classical algorithms. I have used it to train models end to end: forward pass, loss calculation, backpropagation, and weight updates.

EXPLORE THE REPOSITORY ↗
ROLECREATOR
COREC17 + PYTHON API
USED FOREND-TO-END MODEL TRAINING
The actual Bob V2 terminal interface, showing system information, its ASCII robot, quick commands, and the coding prompt
ACTUAL BOB V2 CLILOCAL DEMO SESSION
TEAM-BUILT CO-OP PROTOTYPE / NOT AN OFFICIAL IBM PRODUCT

02.3 / SELECTED PROJECT

BOB V2 — A coding agent built for the terminal.

Bob V2 was a team-built demo for a terminal-first coding workflow. I led the co-op team that made it. The prototype can inspect and edit a repository, run approved commands, review diffs, create commits, start subagents, connect to MCP tools, and control a browser through a local Chrome bridge.

It is experimental co-op work, not an official IBM product.

EXPLORE THE REPOSITORY ↗
ROLETEAM LEAD
CONTEXTIBM CO-OP DEMO
INTERFACEPYTHON CLI / CHROME
01 / SIGN → TEXTONE FRAME.
THREE STATES.

The visual now shows the actual computer-vision sequence instead of reducing a hand to an arbitrary grid of dots.

Clean webcam frame of a hand forming the ASL letter A before computer-vision processing
01CAMERA FRAMERAW INPUT
Actual Roots in Sign webcam output with MediaPipe landmarks and a hand bounding box
02HAND TRACKINGACTUAL REPOSITORY OUTPUT
Tracked hand frame used by the Roots in Sign letter classifier
KERAS CLASSIFIERAPREDICTED LETTER
03CLASSIFICATIONA—Z OUTPUT
02 / SPEECH → SIGNVOICE BECOMES A POSE SEQUENCE.
CAPTUREVOICELIVE MICROPHONE INPUT
WHISPERTEXT
“HOW ARE YOU?”TRANSCRIBED SPEECH
VERIFIED TEXT OUTPUT
LOOKUPGLOSS + POSE
HOWYOU2 POSE SEGMENTS
SIGNED MOTION DATA
Generated signed-video output from the Roots in Sign repository
RENDERED SIGN VIDEO
TWO DIRECTIONS / ONE COMMUNICATION PROTOTYPE

02.4 / SELECTED PROJECT

ROOTS IN SIGN — Two translation paths in one communication system.

For sign to text, Roots in Sign extracts 21 hand landmarks from a webcam frame, flattens their XYZ coordinates, and classifies fingerspelled letters A–Z with a Keras model. In the other direction, it transcribes speech, converts the text to gloss, looks up and joins pose sequences, then renders signed video.

The webcam result and generated sign shown above are real repository outputs.

EXPLORE THE REPOSITORY ↗
ROLEDEVELOPER
STACKMEDIAPIPE / KERAS / WHISPER
STATUSOPEN SOURCE PROTOTYPE

03 / EXPERIENCE

EXPERIENCE.

Engineering, research, product development, and technical leadership—shown through the systems, decisions, and responsibilities behind each title.

IBM
LONGMONT / HYBRIDTHREE ROLES
ONE ENGINEERING PROGRESSION
IBM ↗BOB V2 ↗
01

University Co-op / Cognitive Tooling

Building Unified Chat MCP tooling that helps multiple chatbot surfaces share a dependable way to discover, authenticate, secure, and run tools.

  • Defining shared MCP and chatbot interfaces across standards, integrations, and product boundaries.
  • Working through authentication, security, production engineering, and the less-visible details required to move tooling beyond a demo.
02

AI Development Team Lead

Led the co-op team and hands-on development of Bob V2, a terminal-first coding-agent prototype, from product direction through implementation.

  • Built workflows for repository inspection and edits, approved command execution, diff review, commits, and delegated subagents.
  • Connected MCP tools and a local Chrome bridge while keeping the project clearly framed as experimental co-op work—not an official IBM product.
03

Level 1 Intern

Researched communication and media pain points, then designed an IBM watsonx concept showing how agentic systems could support concrete business outcomes.

  • Translated an open-ended problem into a technical solution plan, prototype direction, and presentation.
  • Worked with an engineering team across development, testing, integration, and client-focused technical communication.
Gray Area Labs
DEC 2025—PRESENTFOUNDER / BUILDER

Gray Area Labs

Building verification-first AI systems and the lower-level infrastructure needed to understand, test, and trust them.

  • Created GrayGate’s Qiskit generation, execution, validation, and targeted repair loop.
  • Built GreyML’s C17 tensor engine, Python API, autograd system, trainable layers, losses, optimizers, and data utilities.
  • Developing evaluation harnesses and developer tools that make claims reproducible instead of merely impressive.
VISIT GRAY AREA LABS ↗GREYML ↗
NSF AI Institute for Student-AI Teaming
JAN—APR 2025RESEARCH INTERN

NSF AI Institute for Student-AI Teaming

Developed multimodal AI work for understanding speech, action, gesture, and shared belief in collaborative learning environments.

  • Fine-tuned a BERT model for shared-belief detection, reaching 83% accuracy.
  • Contributed to spatial AI work using IoT sensing, computer vision, and visual analytics.
  • Presented technical concepts in student-focused settings at CU Boulder.
VISIT iSAT ↗
St. Vrain Valley Schools
DEC 2024—AUG 2025AI LEADERSHIP TEAM MEMBER

AI Leadership Team at St. Vrain Valley Schools

Built applied AI projects and helped turn them into hands-on technology experiences for students across the district.

  • Developed an ASL translator, people counter, and drone-tracking system using machine learning, computer vision, and IoT inputs.
  • Designed AI projects, games, apps, and lessons that helped younger students explore technology through hands-on activities.
  • Worked across prototyping, implementation, testing, demonstration, and technical explanation.
ST. VRAIN VALLEY SCHOOLS ↗

04 / EDUCATION

EDUCATION.

DATA SCIENCE / MACHINE LEARNING / APPLIED MATHEMATICS

Formal study in the models, mathematics, data systems, and computing foundations underneath the work.

01Metropolitan State University of Denver
DENVER, COLORADO

Metropolitan State University of Denver

DOUBLE MAJOR

Current undergraduate work across machine learning, data systems, mathematical modeling, and the foundations needed for advanced AI research.

VISIT INSTITUTION ↗
B.S. DATA SCIENCE & MACHINE LEARNINGB.S. APPLIED MATHEMATICS
EXPECTED MAY 20302030
02
LONGMONT, COLORADO

Front Range Community College

P-TECH / DUAL ENROLLMENT

Completed alongside high school through P-TECH, building a practical foundation in programming, databases, systems, and applied information technology.

VISIT INSTITUTION ↗
A.A.S. COMPUTER INFORMATION SYSTEMS
EXPECTED DECEMBER 20262026
03St. Vrain Valley Schools
LONGMONT, COLORADO

Skyline High School P-TECH

P-TECH PROGRAM

A career-connected STEM pathway combining high school, college coursework, and industry experience through IBM's P-TECH model.

VISIT INSTITUTION ↗
HIGH SCHOOL DIPLOMA
GRADUATED MAY 20262026
A real ASUS ESC4000 G4 2U GPU server chassis
HOME AI LAB / NODE 01REAL CHASSIS REFERENCE / ASUS ESC4000 G4

05 / THE HOME AI LAB

I wanted infrastructure I could understand, control, and share.

An ASUS ESC4000 G4 in my garage gives local inference, experiments, storage, and shared services a physical home. Dual Xeon Gold 6138s, 96 GB of memory, Ubuntu Server, and a GPU build still in progress make the infrastructure part of the research process.

WHY
Privacy, control, learning, access
RUNNING
Local models, experiments, shared services
NEXT
GPU acceleration, better observability

06 / CURRENT

Current work
and contact.

CURRENTLY

Studying data science and machine learning at MSU Denver while continuing at IBM and building GrayArea Labs.

OFF THE CLOCK

Hardware, lifting, reading, design, and taking complicated things apart until they make sense.

LOOKING FOR

Ambitious technical problems, strong collaborators, and work where correctness actually matters.