Skip to content

Workshop

Alongside client work, Kropf Systems develops own systems, prototypes and investigations. They answer specific questions in practice: whether an approach works, where its limits are and what can be learned from it.

Some later become a client project or a product of their own, others remain an experiment.

AI Onboarding Agent

Question

How does an ID card or work permit become a verified record without anyone having to transfer all the details by hand?

Approach

The document is captured with a phone. The agent reads the recognisable details and turns them into structured data. If details are unclear or missing, it asks. Before anything is saved, it shows the captured data for confirmation.

Insight

Most of the time goes not into capturing the details but into chasing missing ones. If the system asks straight away, the record is complete the first time.

Status

Prototype, submitted to the Google Gemini Live Agent Challenge in March 2026 and no longer running. The demo video and the project description are public. Last checked in October 2026.

Watch demo View project

Keeping rulebooks aligned

Question

How does a growing rulebook stay complete and free of contradictions when several documents are constantly changed?

Starting point

With every new version, content can be changed, moved or removed. With several related documents there is the added risk that statements no longer fit together or that the same rule is described differently in several places. Checking by hand becomes more and more laborious as the volume grows.

Approach

A repeatable comparison checks new versions against their predecessors and related documents against each other. It looks for:

  • content that has disappeared
  • contradictory statements
  • duplicates
  • changes between versions

Places it finds are reported as findings, each at its place in the document.

Insight

What gets lost in a revision hardly stands out when reading, because you see what is there and not what is missing. Comparing against the previous version shows exactly that.

Status

Prototype, tested in April 2026. Last checked in October 2026.

Asking your own documents

Question

How do you get a reliable answer from extensive documents of your own without having to search through everything yourself?

Approach

A question is asked in your own words. The answer comes from your own documents, together with the place where it is written. What is not in them is not made up but named as a gap.

The tool is also suited to AI agents. They receive their information with its source instead of searching the documents themselves.

Insight

An answer is only useful once you can see what it rests on. The source makes every answer verifiable, for people and for agents alike.

Status

Prototype, in use since September 2026. AI agents have been using it since October 2026. Last checked in October 2026.

Practice management for appointment-based businesses

Question

How does a business that works with appointments run its whole administration in one place, from the booking to the billing, without online booking, front desk and till drifting apart?

Starting point

In small businesses, appointments, customer data, the till and billing often sit in separate tools. The guest notices the result first: a booking confirmed online does not count at the front desk.

Approach

One system for the whole business: appointments and online booking, rooms and staff, customers and reminders, payment, receipt and till, gift vouchers and reports. Those who need more add billing with health insurers, staff pay statements, products and stock, marketing or further locations.

An online booking is confirmed straight away when the slot is genuinely free. A photographed paper diary is read, and the appointments are ready without anyone typing them in.

Insight

A double-booked appointment costs more than a booking brings in: an annoyed guest and a bad review. That is why it matters that online, front desk and till all show the same picture.

Status

Own system, in development. No business uses it yet. Last checked in October 2026.

Running projects with many people involved

Question

How do you know at any time in a large project what has happened, what is coming up and where things are stuck, without spending hours keeping it up to date?

Approach

Notes, emails and minutes flow in as they arise. From them the system keeps the state of the project: topics and meetings, open points and action items, what has been clarified and what still needs clarifying, decisions and what remains to be decided, commitments and agreements, risks, changes and acceptances, and who is responsible for what.

Anyone who looks in sees at a glance what has happened, what is coming up, what is stuck and who is being waited for. Confidential material stays within the business.

Insight

What happens in a project is in emails and meetings, not in the project tool. A tool that goes there stays current. One that asks people to enter things afterwards does not.

Status

Own system, described and built as a first prototype with made-up data. Not yet in use in any project. Last checked in October 2026.

A similar question?

Do you have a task where it is not yet clear whether or how it can be solved sensibly with technology? Then a limited pilot or an investigation can be a good way in.

Get in touch info@kropfsystems.ch