Role: Web Developer / Technical Artist
Tech Stack: WebGL, TensorFlow.js, Three.js, JavaScript, HTML/CSS
URL: anarki.kylecypher.com
Overview
Anarki is a browser-based application I developed that allows users to generate synthetic training data for machine learning, all while preserving complete user privacy. Designed in a way that practically anyone can use it, Anarki runs entirely locally, meaning that no data ever leaves the user’s device. There is no logging in making the page frictionless for the user.
This tool empowers users to import their own 3D models, configure rendering options, generate thousands of dataset images, and even train machine learning models, it's all directly in the browser with GPU acceleration.
Key Features
-
Privacy-First Architecture:
No servers, no uploads. All computation happens locally in the browser. -
Synthetic Data Generation:
Users can create large-scale image datasets (including segmentation masks) from 3D .glb models using customizable settings like angles, lighting, and backgrounds. -
WebGL + Hardware Acceleration:
Rendering is GPU-accelerated via WebGL, enabling fast generation of complex visual data. -
In-Browser ML Training:
Integrated TensorFlow.js support allows users to train machine learning models directly in the browser and export them for use in their projects. -
Advanced Customization:
Fine-grained control over scene setup for maximum dataset variability.
The Process
-
Import 3D Models
Upload .glb files, they remain fully on-device. - Configure & Render
Set lighting, camera angles, and backgrounds. Generate thousands of image-label pairs in seconds.
-
Train & Export
Train ML models locally and export both datasets and model files for use elsewhere.
Outcome
Anarki serves as a fully local, secure, and efficient alternative to cloud-based synthetic data generation tools. It’s ideal for developers and researchers who need total control over sensitive training data, from creation to deployment, without sacrificing performance or flexibility.