top of page
Abstrakter Hintergrund

Recipe scan

Digitize doctor's prescriptions quickly and reliably

3 AI Developers

Project duration
9 months

Product success
More than 4 million recipes per month

Reading out doctor's prescriptions 

For the scanning line of the ARZ Darmstadt, we have developed a high-performance software solution based on our AI technology for reading out sample 16 recipes. With this AI solution, we were able to reduce the error rate and thus the manual rechecking by more than 30% - and this at a readout speed of less than 0.02s per prescription. With 4 million prescriptions per month, we were thus able to help ARZ Darmstadt to make the processing of doctors' prescriptions much more cost-efficient. What's more, this internal project has now become a product that we would like to make available to third parties together with the ARZ.

Our partner: 


Our services: Consulting, AI development, deployment, continuous development and support

Technologies used : CNN, RNN, Beam Search, REST API, Docker

Reiner Stürmer.jpeg

"We are very happy about the cooperation with the kaitos team. Thanks to kaitos' AI-based OCR, we were able to reduce the already low error rate in the automatic processing of prescriptions by a third compared to our previous solution, and thus significantly reduce our effort in the manual post-processing. Thus, kaitos AI solution is a central component for us to continue to operate the billing of prescriptions economically with very high performance and quality in the future.”


Reiner Stürmer, Technical Lead of ARZ Darmstadt

Recognition rate improved by a third

0.02s per recipe (on an NVIDIA RTX 3090)

The data is processed in compliance with the DSGVO under the highest standards

Abstrakter Planet

And how does it work?

The prescriptions are scanned using a scanning path and are thus made available for reading the individual fields.

 Create scan

The recipe scans are then sent to the AI model deployed locally via Docker container, which can be addressed directly via API. The model runs in parallel on several Tesla GPUs and can therefore be scaled accordingly and designed to be fail-safe.

Using the model

Within the AI, the recipe is broken down into the individual fields to be read and these are read out - even for handwritten fields reliably and quickly. The data is then returned to the client structured as json.

Processing by the AI service

The data can then be freely processed on the client side - and all within fractions of a second per recipe.

Use of the result

Dorotheenstraße 26, 48145 Münster 

+49 (0) 251 37 96 83 83

For more information

or information about our services

Seal of ApprovalTransparent (1).png
  • LinkedIn
bottom of page