Home
Personal projects, 7
Esc

projects

Things I build after hours: tools for AI agents, small utilities for Chișinău, and languages I learn by building.

All 7 projects

Markdown is a poor format for working with coding agents: it costs tokens, and agents skip steps or make things up. cartograf replaces it with a generated map of the code and a small language for flows, served to the agent over MCP.

How it works

  1. 01A generator builds a map of the codebase and keeps it up to date.
  2. 02Flows are written in a small DSL instead of prose, so steps cannot be skipped.
  3. 03An MCP server hands both to the agent; agents write and maintain these files themselves.
Stage
  1. idea
  2. prototype
  3. building
  4. live
Stack
  • MCP
  • DSL
  • Code map
Links
GitHublink coming

A small task board for the AI agents I work with. Tasks, subtasks, comments and progress live in a floating window, and everything, including client conversations and requests, goes into one database I can ask questions.

How it works

  1. 01Agents create and update tasks through an MCP server.
  2. 02A floating desktop window shows tasks, subtasks, comments and when each was done.
  3. 03Vector search answers questions like “when was this built, and based on what?”.
Stage
  1. idea
  2. prototype
  3. building
  4. live
Stack
  • MCP
  • Tauri
  • Web
  • Vector search
Links
GitHublink coming

Every half hour it reads the outage announcements for water, power, gas and heating in Chișinău and tells the people on the affected streets, before the tap runs dry.

How it works

  1. 01Checks the utility operators’ sites every 30 minutes.
  2. 02Matches announced streets to subscribers.
  3. 03API first; the app or Telegram bot on top comes later.
Stage
  1. idea
  2. prototype
  3. building
  4. live
Stack
  • Elixir
  • Phoenix
  • Oban
Links
GitHublink coming

A small Telegram bot for Chișinău: send where you are and the trolleybus you want, and it tells you how long until it comes. Plain arrival time is only the start.

How it works

  1. 01You send your location and a route number.
  2. 02The bot finds the nearest stop and the next vehicle.
  3. 03It replies with the time until arrival.
Stage
  1. idea
  2. prototype
  3. building
  4. live
Stack
  • Telegram
Links
GitHublink coming

A journey planner for Chișinău public transport built on RAPTOR, a fast round-based routing algorithm. It is also how I learn OCaml: by building something real, step by step.

How it works

  1. 01Timetables are loaded into compact arrays.
  2. 02RAPTOR scans routes round by round, one round per transfer.
  3. 03The same OCaml code compiles natively and for the browser.
Stage
  1. idea
  2. prototype
  3. building
  4. live
Stack
  • OCaml
  • RAPTOR
Links
GitHublink coming

A browser of my own: fast, light on memory, safe, and private by default. Extensible like Hammerspoon through a scripting language, with a built-in key store where secrets are never kept in readable form.

How it works

  1. 01A small modular core, with pluggable components.
  2. 02Scripting for everything you would want to automate.
  3. 03Built on macOS first.
Stage
  1. idea
  2. prototype
  3. building
  4. live
Stack
  • macOS
  • Scripting
Links
GitHublink coming

My travel photos, one city at a time, in a Swiss poster style. Each city gets its own colour, and you see the photo through the letters of its name.

How it works

  1. 01An index of cities, as a list or as posters.
  2. 02Each city opens with a colour wipe.
  3. 03Photos open as posters you could print.
Stage
  1. idea
  2. prototype
  3. building
  4. live
Stack
  • Design
  • Photography
Links
Open drumuri ↗