Skip to main content

Architecture & constructionFrom our projects · Own product

Who does what by when?

After every site meeting someone digs through the notes to find out who took on what. vera. listens in the meeting, even in Swiss German, writes the minutes and carries the action items forward, and before anything goes out, you approve.

Anonymised data

Starting point.

The question came from an architecture firm. Common AI tools do write minutes, but not in your own template, and the action items start from scratch at every meeting. People speak Swiss German, often mixed with French or English.

Whoever writes the minutes then spends a while with the notes: who committed to what, by when, and what was already open last time?

That is how vera. came about, a Kallysoft product of its own. We use it day to day in our own team.

Who has taken on what by when, and what has been done since the last meeting?

What we built.

A web portal and a small desktop app for recording. The AI listens and writes the draft, fixed rules check afterwards.

  1. Listening to who says what AIRecording happens in the meeting room or in the online call, or you upload the recording. The software separates the voices and passes them to speech recognition trained on Swiss German.
  2. Minutes with sources AIThe AI cleans up the transcript with the project's glossary and writes decisions, open questions and action items from it. Every statement points to the place where it was said. What was not clearly understood stays as "unintelligible".
  3. Action items that do not get lost calculatedEvery action item keeps its number across all meetings, with a history and a quote for each step. What someone changed by hand is not overwritten by any later run.
  4. In the office's template calculatedThe minutes are read and corrected in the portal. Only then is the Word file created in the template the office specifies. Sign-in is with the Microsoft 365 account.

What the AI does not do.

  • It approves nothing. The minutes stay a draft until a person approves them. No file is created before then.
  • It does not tick things off silently. An action item is only done if it was said in the meeting. What nobody mentions stays open.
  • It does not guess. Names come only from the participant list, and the software calculates dates itself. Unclear deadlines such as "end of month" and speaker names are confirmed by a person. And Swiss German remains difficult: not every dialect works equally well.

What we learned.

It did not get better through a new model.The obvious step would have been to swap the speech recognition. But the minutes got better through rules around it: names only from the participant list, dates from the software instead of the AI, and every classification needs a source.
Licence first, then the recommendation.Once we recommended a Swiss German model that may not be used commercially at all. We noticed and corrected it. Since then the licence is right at the front of our model overview.
Real conversations stay out.To try out an integration a real internal meeting was ready, and it would have been convenient. We made up examples instead. Real conversations do not belong in a test.

How do your minutes come about today?

Tell us how your meetings run and what your template looks like. In a conversation we will see whether vera. suits you, run by you or by us. No obligation, of course.

Send a project enquiry

reCAPTCHA *
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.

Lettenweg 40
4123 Allschwil

Tel: +41 (0)61 544 05 00
Support Tel: +41 (0)61 539 10 80

Email: This email address is being protected from spambots. You need JavaScript enabled to view it.

© Kallysoft Informatik AG All rights reserved.

Made with in Switzerland

Swiss Made Software logo
Swiss digital services logo
Swiss Hosting logo