20 Mar 2026
This week, we reached an initial conclusion with a proof of concept we have been working on for the last month or so, which we have lovingly titled “The Form Filler”.
The goal was to create a tool that helps SENCOs with a repetitive and notoriously tedious task: filling out forms. Our research highlighted that SENCOs often spend a huge amount of their time essentially copying information from one form to another. Although the information required is broadly similar, there are sufficient differences to warrant specialist oversight. Beyond the obvious soul-destroying drudgery, this is a problem because it consumes precious SENDCO time that could be put to much better use elsewhere.
The tool works in three stages. First, the form is converted into images, and a large language model (LLM) is used to understand the form structure. Second, the required answers are generated using anonymised pupil data (we talked about anonymisation in last week’s update). Third and finally, the form is completed using the generated answers, and can then be downloaded ready for submission.
What worked well? The LLM was able to understand the form structure and the answer generation was fast and sufficiently accurate.
What needs work? On occasion, the LLM misclassified form guidance as something that needed to be answered. We also found that many of the sample forms we were testing the tool with asked whether there were any changes to the pupil, for example, their support needs. For the tool to answer such questions effectively, it would need access to historical records, not just a single snapshot.
We wish you a fantastic weekend.
