sefaonder.dev
// WORKBENCH / VITRINE / SELECTED_WORKS

Selected Works & Systems Architecture

Things I've built or am building. Mostly local-first tools and client-side architectures where the data stays on the device.

2 Featured Work
100% On-Device Storage
0ms Offline Latency
Open Source Public Source Available
Production 2026 Creator & Core Maintainer
Interactive preview screenshot of Local Focus
Local-First & CRDTs
Zoom
Architecture Strategy

Fully client-side. Pinia for reactive state, IndexedDB via idb for persistence, Service Worker for offline caching (vite-plugin-pwa).

Key Engineering Trade-off

Zero server means no cross-device sync. Users export and import JSON files manually instead.

Performance Metric

Instant 0ms timer loop & mutation resolution. Fully functional without internet.

Storage Subsystem

Browser IndexedDB (via idb abstraction)

Local Focus

Offline-first Pomodoro timer and task tracker. All data stays in the browser.

The constraint was simple: no server, no account, no tracking. A Pomodoro timer with task-linked sessions and productivity stats, backed entirely by IndexedDB. Works offline out of the box via PWA caching. Backup and restore uses portable JSON file exports.

0% (Local-Only) Cloud Dependence
IndexedDB (idb) Storage Engine
100% PWA Offline Ready
Zero Tracking Privacy Policy
Beta 2026 Creator & Core Maintainer
Interactive preview screenshot of Digital Twin Concept Demo
Systems & 3D Simulation
Zoom
Architecture Strategy

Three.js scene graph for 3D conveyor kinematics, WebGL render target for virtual camera capture, and WebAssembly SIMD pipeline for hierarchical workpiece isolation and defect detection.

Key Engineering Trade-off

Running WebAssembly computer vision on the main thread alongside 3D rendering causes frame drops during complex contour extraction, requiring throttled sensor acquisition.

Performance Metric

Sub-15ms edge inference per frame. Zero external API calls.

Storage Subsystem

In-memory ring buffer for recent defect snapshots

Digital Twin Concept Demo

Browser-based 3D industrial automation simulation with OpenCV WebAssembly defect inspection.

The bottleneck in industrial automation prototyping is physical hardware availability and testing sensor edge cases safely. Built a browser-based kinematic twin combining Three.js physics with an emulated telecentric industrial camera. Inspection runs client-side through OpenCV.js compiled to WebAssembly with SIMD, backed by an autonomous pixel scanner fallback. Continuous canvas frame captures at 30 FPS caused garbage collection spikes, which required pre-allocating fixed typed arrays instead of allocating new image buffers on every tick.

60 FPS WebGL Kinematics Loop
OpenCV.js WASM Vision Pipeline
Telecentric 256px Camera Sensor
<15ms / frame Edge Inference