Talking your way out of paperwork

Let’s face it: social workers don’t choose their career for the love of paperwork. They become social workers because they want to help people. Yet every home visit ends with the same unavoidable task: a long, structured, multi-page document describing the financial, social and personal situation of the person they met. Even for native speakers this is demanding work.

For many of Stad Mechelen’s social workers, who come from diverse backgrounds and sometimes speak Dutch as a second language, report writing had turned into a time-consuming chore. In fact, all those reports add up to hundreds of pages of notes and hundreds of hours of work. Surely, this can be done better and more efficiently.

What if writing a report could be as simple as having a conversation?

The challenge

Every day, Mechelen’s social workers visit people who rely on support for housing, income, employment, or day-to-day stability. During these conversations, they gather a huge amount of information: family composition, financial details, medical situations, levels of self-sufficiency, and dozens of other small but important contextual details.

All that knowledge then has to be turned into two structured reports (an internal conversation report, and a formal report for the city administration) that follow a strict template. These reports easily run to two or three pages each. They must be complete, consistent, and written in clear, professional language.

There were four key challenges:

1. For some of the social workers, Dutch isn’t their first language.
2. Writing reports took a very long time.
3. Some people have a more fluent writing style than others.-
4. Important data was not always captured consistently

Stad Mechelen was looking for a solution that would support their social workers without turning them into data-entry robots.

Our speech-to-report tool

Working closely with Stad Mechelen, we came up with a simple yet transformative idea: let social workers speak freely and have AI handle the rest. The result is a fully web-based application that converts natural speech into clean, structured, ready-to-use reports.

Step 1: Speak

The social worker logs into the browser-based tool, clicks ‘Record’ and simply starts speaking. They can talk the way they normally would: in dialect, with local jargon, mixing shorthand, abbreviations and informal phrasing.

Step 2: Transcribe

Using Whisper (OpenAI), we convert the spoken narrative into text. In early tests Whisper already achieved around 90 percent accuracy. Dialect, accents and jargon were the remaining challenges.

To close that final gap, we trained a lightweight correction agent that fixes recurring errors and resolves domain specific abbreviations. This pushed accuracy to an impressive 97 to 98 percent, even with rapid speech or imperfect Dutch.

Step 3: Structure

The real magic happens in this final step with the second AI agent. It analyzes the transcript and organizes the content according to Stad Mechelen’s reporting template. It recognizes personal details, financial context, household composition, and other required categories. Importantly, the tool never invents information or makes recommendations: it simply puts the right pieces in the right boxes.

Once the report is generated, the social worker can copy-paste the output directly into their existing system. No data is stored on any server (nothing is logged) and a simple browser refresh clears the entire session.

When dealing with sensitive information you cannot compromise. The entire solution was built with GDPR and the EU AI Act firmly in mind. The tool doesn’t keep any form of information: everything you say is processed instantly, and only lives in your browser while you’re using it. Nothing is stored, nothing is saved, and nothing leaves your computer. The AI doesn’t make decisions or assumptions, it just helps sort out your notes. You stay in full control, always.

The future

In the near future, this tool will help transform the way reports are created. By making speaking feel more natural and accessible than writing, reporting will become smoother, more consistent, and higher in quality. Most importantly, social workers will be able to spend more time supporting citizens and less time staring at a blinking cursor.

This is just a glimpse of what becomes possible when AI fits seamlessly into real-world workflows. Curious what it could do for your team or organisation? Don’t hesitate to reach out.