Learning AI by building

I build small, useful software with AI, one step at a time.

Open-source apps, a private AI second brain, and a long career in network security behind them.

  • Open source
  • No tracking
  • No accounts
  • Privacy-first
Beeline bookmark dashboard in dark mode, shown with demo data
Beeline, a bookmark dashboard (demo data)

What I do

Build it, test it, write it down

  1. 01

    Build small, useful apps

    An SSH client, a bookmark dashboard and a family calendar. Free, open source, no accounts, no tracking.

  2. 02

    Test AI on real work

    A private AI second brain with local and cloud models, agents and voice. I use it to build infrastructure automation, web apps and a mobile app.

  3. 03

    Write down what I learn

    The Daily Calf on Substack: notes on learning AI by building real things, slowly.

Selected work

Real projects, published for real

Built with AI assistance. The apps are tested and published through the normal channels; the second brain stays private.

ssh4ai with a server list on the left and a terminal session open, shown with demo data

SSH client

ssh4ai

A free SSH client for Windows, plus a self-hosted web version that runs in Docker. Save your servers once and open them in tabs.

  • Windows
  • Docker
Beeline in dark mode: a grid of bookmark groups such as Work and Learning, shown with demo data

Bookmark dashboard

Beeline

A visual bookmark dashboard that replaces the New Tab page, with groups, tags and favourites. Your data stays on your computer.

  • Chrome
  • Brave
  • Firefox
  • TypeScript
  • React
  • Vite
  • IndexedDB
Famabria month view for November with a day panel and a new-note form, shown with demo data

Family calendar

Famabria

A shared family calendar for notes and reminders. End-to-end encrypted, no accounts, works offline.

  • Browser extension
  • Web app
  • Android
Diagram of how a note travels from a Windows PC through a private Git server to search and backup

Private knowledge base

AI second brain

A private knowledge base that several AI agents search with retrieval-augmented generation, using local and cloud models, with voice input and spoken answers. I use it to build and test the other projects.

  • Self-hosted
  • RAG
  • Local and cloud LLMs
  • Agents
  • Voice

Capabilities

What I can ship

  • Browser extensions

    Chrome and Firefox builds from one codebase.

  • Android and Windows apps

    Shared through the Microsoft Store and GitHub.

  • Docker and CI publishing

    Multi-arch images and automated releases.

  • Store submissions

    Submissions and releases for the Chrome, Firefox and Microsoft stores.

  • Security checks

    Secret scans, dependency audits and code scanning.

  • Privacy-first, offline-first

    Data stays with the user; apps keep working offline.

Stack & approach

How the work is done, and by whom

Built to be checked

  • Source code is public on GitHub.
  • Automated checks and releases in CI, including secret scans, dependency audits and code scanning.
  • Accessibility checks on Beeline, in dark and light themes.
  • No tracking in the apps or on this site, and no cookies on this site.
  • TypeScript
  • React
  • Vite
  • Docker
  • Manifest V3
  • IndexedDB

AI-assisted, and said openly

I set the goals, make the product decisions and test the result. AI writes and reviews much of the code. That split is the point of the exercise: learning where AI helps and where a person has to decide.

Before this, 25+ years in cybersecurity and networking pre-sales and solutions, starting as a developer.

Contact

Questions, bug reports or an idea for an app?

Write to me, open an issue on GitHub, or follow along on Substack.