Loading Workspace
sys.init: initializing tensor cores...
Loading Workspace
sys.init: initializing tensor cores...
Real neural networks, trained in Python, exported and executing client-side via TensorFlow.js — zero servers, sub-2ms latency.
Featured System
A 1D Convolutional Neural Network trained on environmental telemetry data. Exported from Google Colab, executing entirely in the browser via TensorFlow.js.
Model Type
Conv1D
Input Shape
[15, 1]
Train Accuracy
89.1%
Val Accuracy
71.9%
Research Experiments
Additional machine learning systems built across different domains — from multi-agent orchestration to tabular regression and clinical EDA.
Multi-Agent AI
Multi-agent system using CrewAI orchestrating specialized LLM agents to autonomously design, develop, and audit retro arcade games in Pygame.
Tabular Regression
Benchmarking SVR, Random Forest, Decision Tree, and KNN models on delivery logistics and California real estate pricing datasets.
Clinical EDA
Exploratory analysis of physiological attributes influencing coronary artery blockage. Visual correlation mapping of heart attack risk indicators.
Association Mining
Interactive graph of Apriori-mined association rules from a deforestation and transaction dataset. 42 rules visualized with support/confidence/lift filtering.
System Architecture
The complete data lifecycle from raw ingestion to browser inference — each stage animated and interactive.
Curriculum
18 notebooks across Foundations, Classical ML, Statistics, Deep Learning, and Advanced Analytics — structured as an interactive progression map.
Engineering Decisions
Key architectural tradeoffs and design choices that make this a production-quality engineering showcase.
Running TF.js in the browser eliminates server round-trip latency. Predictions happen in <2ms vs ~150ms for an HTTP request — and at zero server cost.
TensorFlow.js and canvas-heavy components are loaded via `next/dynamic` with `ssr: false`, ensuring clean hydration and optimal bundle splitting.
Models trained in Google Colab (Keras) are exported to TensorFlow.js format and served as static JSON from `/public/models`, enabling zero-dependency inference.
I'm actively looking for ML engineering and software engineering roles. Let's talk.