Flagship
Atlas
Problem: instructors need faster prep, grading support, and early risk
signals without surrendering judgment to automation.
Approach: academic intelligence layer where AI drafts lessons and
assessment feedback; humans review and approve. Start from uploaded course materials;
LMS connect is an upgrade path.
Outcome: live instructor pilot with private workspaces, materials,
Prepare, and Assessment Intelligence.
React
Supabase
EdTech
LLM workflows
Computer vision
Pineapple Counter
Problem: count produce on a moving conveyor without cloud-dependent
vision infrastructure.
Approach: React Native app with on-device object detection (TensorFlow
Lite), virtual count line, duplicate suppression, and local history.
Outcome: iOS/Android prototype for real-time counting; demo available
on request (repo not public yet).
React Native
TFLite
On-device ML
CV
Applied probability
PitchPrice
Problem: soccer forecasts are easy to present as picks and hard to
evaluate honestly.
Approach: Dixon–Coles modeling, chronological validation,
calibration / scoring rules, and model-vs-market (no-vig) comparison in a Streamlit
product surface.
Outcome: offline-reproducible diagnostics app; bundled demo data is
synthetic by design.
Python
Streamlit
scikit-learn
Calibration
Theory → tools
TDA · Fractals · Interactive math
Public repositories that make geometric and topological ideas computable and teachable:
topological data analysis experiments, Mandelbrot/Julia exploration, and standalone
interactive math experiences for concept learning.
TDA
Complex dynamics
Math education
Precursor
Learning Analytics MVP
Earlier full-stack experiment (React + Node + OpenAI) for role-based tutoring modes and
analytics dashboards. Design lessons and AI workflow patterns carried forward into
Atlas.
React
Node
OpenAI