$ ls blog/

Notes from the trenches: AI workflows, cinema tech, Bitcoin, projects

⚡ AI Radar: my daily AI and robotics news feed

AI LLM Claude Code Cinema SNMP Node-RED Bitcoin Lightning Self-hosted Projects
2026-06-195 min read

Vibecoding is 80% setup. Here's mine, step by step.

Most people open an agent, type „build me an app“ and wait for magic. I do it the other way around: guardrails, plan mode, multi-model plan review and parallel agents are in place before a line is written.

AIVibecodingClaude Code
2026-06-197 min read

Claude Code vs Cursor vs Codex: which to use when (2026)

I use Claude Code, Cursor and Codex every day. They solve different problems. Here is when each one actually shines and how I combine them in 2026.

Claude CodeCursorCodex
2026-06-196 min read

5 things that burned me in agentic AI development (and how I do it now)

Five real failure modes from daily agentic AI coding with Claude Code, Cursor and Codex, and the concrete fixes I actually use now.

Agentic AIClaude CodeWorkflow
2026-06-196 min read

Self-hosted vs cloud for AI development: why I keep my infra on my own VPS

Why I run my own VPS, Postgres and Redis for AI dev instead of a cloud platform. The honest tradeoffs: control, cost, privacy, and the ops burden.

Self-hostingInfrastructureAI
2026-06-196 min read

Self-hosted Bitcoin node: why sovereignty isn't just a buzzword

Why I run my own Bitcoin and Lightning node, what financial and data sovereignty mean in practice, and the trust model nobody tells you about.

BitcoinSelf-hostingPrivacy
2026-06-196 min read

I built my own offline speech-to-text for macOS, here is why

Dictate: hold a hotkey, speak, release, text appears. Whisper runs locally on macOS, nothing leaves the machine. Why local-first dictation matters.

ProjectsPrivacymacOS
2026-04-104 min read

A local model for vibecoding: Qwen 3.5 Coder on an M5 Max

Which model I run locally for coding: Qwen 3.5 Coder on an M5 Max. Surprisingly usable, and ideal for sensitive or production work I do not want to send to an online LLM.

Local LLMPrivacyVibecoding
2026-01-055 min read

PF 2026: building a New Year's browser game in a single sprint

A New Year's greeting as a browser game: calibrate a cinema projector, race the clock, get an AI video at the end. Idea to deploy in one sprint.

ProjectsBrowser GameCinema
2025-11-105 min read

Let the LLM draw it: generating Mermaid diagrams with AI

How I use an LLM to turn a description or existing code into Mermaid diagrams: flowcharts, sequence, ER, state. Diagrams as text that render anywhere.

MermaidAIDocs
2025-11-025 min read

I taught a script to draw photos with letters: a weekend ASCII art toy

A weekend hack: turn photos into ASCII art with custom letter-shaped masks. Python, Gradio and PIL, live at ascii.ondra-vlasek.cz. Why small toys matter.

ProjectsPythonWeekend Hack
2025-11-016 min read

Vibecoding Three.js in 2025: when AI hit the wall of 3D

Autumn 2025, almost nobody used AI for 3D in the browser. I did anyway. Honest notes on vibecoding Three.js: where models shone and where they fell apart.

VibecodingThree.jsWebGL
2025-10-155 min read

An interactive 3D LED wall visualizer (and a 2025 model benchmark)

For a client buying an LED wall I built an interactive 3D demo with real dimensions and pixel pitch, then got carried away with a reference figure, VR mode and Telegram. A drawer project and an October 2025 model benchmark.

ProjectsWebGLFrom the drawer
2025-08-156 min read

I built my own Lightning wallet in a summer with Cursor

How I vibecoded a non-custodial Bitcoin Lightning wallet (iOS, Android, web) on top of LNbits in one summer with Cursor. Keys local, control mine.

BitcoinLightningVibecoding
2024-02-155 min read

Fabric was right early: prompts as tooling, not chat

Two years ago Fabric treated prompts as reusable, version-controlled CLI tooling. It quietly predicted CLAUDE.md, skills and prompt libraries.

AIToolingFrom the drawer
2023-05-154 min read

AI Shell: the quiet predecessor of vibecoding

Two years before Cursor and Claude Code, a humble CLI already let you say what you wanted and get an executable command back. A from-the-drawer look back.

AI ShellVibecodingCLI
2023-02-155 min read

My first vibecoding: ChatGPT, copy-paste and a model that could not add up

My first vibecoding at the dawn of ChatGPT: copy-paste, Stack Overflow and a model that could not even add up. From day one of ChatGPT Plus to today's agents.

AIVibecodingFrom the drawer