Uses

The stack I ship in every day.

• React 19 + TypeScript the day-job stack — typed components end to end
• Tailwind CSS v4 utility-first styling, design tokens via @theme
• shadcn/ui on Base UI component registry, themed to my own violet tokens
• Vite the bundler under everything — fast HMR
• Next.js app-router React for the console — SSR to the edge
• TanStack Start, Query, Router, Table, Form — data, routing and tables
• Zustand small, unceremonious client state
• Radix + Base UI headless primitives under the component kit
• Zod + Axios schema-validated data over typed HTTP
• Cloudflare Workers where the apps deploy — edge runtime
• Turborepo + pnpm the monorepo that holds every app and the UI kit
• Voximplant + WebRTC voice AI and in-app video — the call center and video calling

The editor, API and comms tools I live in.

• VS Code the editor
• Postman poking APIs before wiring them up
• Slack where the team coordinates
• Git + GitHub version control and reviews
• Claude Code pair-builds with me daily

What this page you’re reading is built with — static, zero JS by default.

• Astro static output, ships zero JS unless a component asks for it
• MDX blog posts — markdown with components when needed
• GitHub Pages + Actions push to main → build → deploy, no manual step
• satori → resvg build-time OG images seeded from the knowledge-graph motif
• Shiki dual-theme syntax highlighting in posts

The second brain, and what keeps it current.

• Obsidian the vault — plain markdown, git-synced
• Claude Code operates the vault and builds with me daily
• MCP how AI clients reach the vault as tools
• Python vault tooling — health checks, retrieval evals

What I reach for across models, providers and agent tooling.

• Claude my daily driver for building and writing
• Many providers OpenAI, Google, DeepSeek, Copilot, OpenRouter — matched to the task
• Ollama local models on my own box — Llama, Gemma
• Agent CLIs Claude Code, OpenCode, Kilo Code, OpenCLAW and friends
• MCP + skills tools, agent skills and workflows wired into daily work
• n8n (self-hosted) automation on my own DigitalOcean box
• Context engineering the real skill — the right context in front of the model, every time

How it all runs — the habits, not the hardware.

• Git-synced vault one knowledge base, every surface, synced over git
• AI-first notes written for a machine to read first — preamble, links, sources
• Scheduled agents jobs that maintain the vault on a cadence — nothing deleted, only added
• Mobile capture quick thoughts on the go via a Telegram bot, processed later
• Retrieval eval search quality is a number I can move, not a vibe