Simplifying dog tax registration using Amazon Bedrock with bol

Headquartered in Munich, Germany, bol designs specialized software solutions that accelerate digital transformation in the public sector. As AI matured, the company saw an opportunity to use advanced technology to reduce the administrative burden of processing and submitting government forms online.

This idea sparked a collaboration with a city administration that’s one of bol’s long-standing clients. The team identified the city’s dog tax registration as a suitable use case for testing a public, self-service AI solution. However, the software provider and the city administration needed to comply with data sovereignty regulations while integrating large language models (LLMs) into the process.

“Compliance is a main priority for us because we work for public services,” says Falk Merten, head of infrastructure at bol. “In Germany, we have strong restrictions on how we handle data, and AWS is the right choice in terms of available services and data protection.”

The software provider envisions its solution as more than a standard form assistant. The company wants FINO to include language translation, automated document verification, and other intelligent capabilities to simplify online registration for German and non-German speakers. By implementing Amazon Bedrock to build generative AI applications and agents at production scale, bol can access and experiment with hundreds of LLMs.

To power FINO’s capabilities, bol uses Amazon Bedrock AgentCore, an agentic service for building, deploying, and operating effective agents securely at scale. By using AI, the company provides citizens with explanations in their native language. “Users can ask FINO questions if they’re having trouble understanding any part of the form,” says Karsten Simons, cofounder and head of business development for AI at bol. “By retrieving data from the city’s knowledge base, the assistant can produce relevant answers.”

The solution also guides citizens through follow-up processes, such as muzzle requirements and proof of liability insurance. “FINO gives not only responses to applicants’ requests but also step-by-step guidance,” says Simons.

As part of the registration process, the city requires applicants to report their dog’s breed. Using AI image recognition, FINO can analyze a dog’s photo, identify the breed, and determine whether it’s on the list of dangerous breeds. Automatically adding this information to the registration form, the solution helps improve the city administration’s data quality.

The company runs its workloads in the AWS Frankfurt Region so that data remains within German and European regulatory boundaries. Using AWS managed services, bol benefits from built-in security features, such as data encryption and access control, without needing to expand its infrastructure team. “We have a small team of software developers, so we rely on AWS managed services to quickly launch new products,” says Merten.

Piloting FINO together, bol and the city administration fine-tuned the assistant’s responses according to users’ feedback. Powered by AI image recognition, the solution identifies breeds with more than 90 percent accuracy and an average response time of 5 seconds.

Using FINO’s architectural pattern, bol can transform HTML forms into intelligent assistants, paving the way for innovation in the public sector. Because the solution functions as a modular web component, municipalities can incorporate it into existing HTML-based services without replacing current systems. This means that other agencies can replicate and scale the solution to suit their regulatory and operational requirements.

Moving forward, bol and the city administration aim to collaborate to develop an assistant for accelerating other kinds of applications. “The city is testing internal LLM chatbots, but this will be the first time that it has offered an AI service for administrative processes to its citizens,” says Simons. “By validating that a form is complete, FINO reduces the need for human intervention and follow-ups, saving time and minimizing errors.”