🚀 Meet Zotra

Posted on In
Zotra, workflow studio for banks

I took part in the Meta Global AI Developer Hackathon 2026 as an indie dev, and I built Zotra: an AI workflow studio that turns a bank’s credit policy (a PDF) into a tested approval app.

Why “Zotra”?

Time for another little Malagasy lesson (the last one was Toko).

Zotra means “path”, “route” or “direction”. We use it in many contexts. For example:

  • Tambajotra: a group of paths, so a network.
  • Zotra Nasionaly: a national road.
  • Miara-mizotra: walking together.

A workflow is a route toward an end, a path to a destination. Just like a zotra. So the name was obvious.

I like to name the tools I build in Malagasy, because I find the language rich in meaning. I’m Malagasy, but it still surprises me how little I know about my own language. Some words feel elegant, even poetic, once you understand them.

Why did I build Zotra?

A bank writes its credit procedures in PDF policies. To turn one policy into a workflow, business and IT teams meet for weeks. And at the end, nobody can easily say which step comes from which rule of the text.

I saw this first-hand, first as a developer, then as a PMO in a bank. We did the same exercise over and over again. The pattern is well known. But even with layers of abstraction, the process stays demanding for the developers, the testers and everyone else involved. Tweaking takes time. The back-and-forth takes time. In a bank, people are rarely available. By the end, days and weeks have passed.

Zotra is my attempt to fix this for those who still struggle with it. In my head, it’s almost a help sent to my past self. I made it open source so that anyone can use it, to update old policies or to create new ones. It can save a lot of time, especially for new banks that try to ship fast to keep up with the competition.

What it does

  1. Read the policy: drop a PDF. The AI drafts the steps, roles, deadlines, documents and rules. Each step cites its paragraph in the PDF.
  2. Shape the process: edit the diagram by hand, or ask the AI assistant in plain words.
  3. Test it: play a case as any role. The AI can explain a failed check and write the test cases you missed.
  4. Ship it: freeze and publish. Each role gets its own app, in the bank’s colors.
  5. Keep the policy up to date: after a change, the AI rewrites the affected paragraphs for the bank to review.

The AI drafts. People decide.

On a live case, the AI never decides. Only the rules and the people do. Nothing from the AI is saved until a person accepts it.

This matters to me. A tool does not think for itself: what it does is decided by the one who holds it. I wrote that about the vakana, and it is still true for AI. In an older post, I said that AI is like a car. Zotra is the road. The bank still chooses where to go.

In Toko, I promised to open the cooking pot. Here is the first dish. More will come, so keep your seat at the Avara-patana and stay warm near the hearth.