bymdp.dev

Analytical tools that keep themselves up to date.

Projects at the intersection of finance, foreign trade and open data on Russia and Japan. Each one takes its figures from primary sources, checks them before publication and reports on its own when something changes.

Projects

3 live · 1 published · 1 on request

Platform

how it works

One server and a few agents on a schedule

All the projects live on a single machine. Scripts and Claude Code agents watch over data freshness and service health. When a project loses a data source, they notice it themselves and prepare a fix — while the decision to deploy stays with a human.

  • Perimeter

    Not a single open port

    The sites reach the outside world through a tunnel: the server accepts no incoming connections at all.

  • Monitoring

    A check every ten minutes

    The scanner looks not at whether the process is alive but at whether the data is fresh: a parser that has quietly stopped working is the main failure of such projects.

  • Repair

    An executor and a skeptic

    One agent fixes the problem in a separate branch, and a second one checks its work. A human merges and deploys — after a message in Telegram.

  • Savings

    The model is for judgment only

    The watchdog, the task dispatcher, the nightly backups and the tests are plain scripts. The model is brought in only where reasoning cannot be avoided.