Introduction

Onboarding a new member of staff, such as a Teaching Assistant (TA) or cover teacher, involves a significant handover of information, particularly when supporting pupils with Special Educational Needs (SEN) or Social, Emotional, and Mental Health (SEMH) needs. Crucial data—like behaviour incident reports and exclusion timelines—can be spread across multiple paper files and digital documents, making it challenging for new staff to get up-to-speed quickly.

But what if a new TA could ask a simple, direct question and receive a transparent, cited summary of a pupil’s background in minutes?

This post details a powerful prototype demonstrated by Mr Vullo, using a local Large Language Model (LLM) in a school setting to instantly compile essential pupil information, ensuring new staff are equipped to provide effective support from day one.

The Challenge of Onboarding New Staff in Schools

In education, a pupil’s success often relies on the continuity of care and support, which can be disrupted when a new staff member joins the team. Accessing and synthesising relevant documentation is time-consuming. Schools need a solution that is both efficient and robustly compliant with data privacy, especially when handling sensitive pupil data.

The prototype addresses this by showing how a local LLM can streamline the retrieval and analysis of fragmented data, turning a lengthy reading process into an immediate, actionable summary.

Introducing the Local LLM Prototype: A Digital Teaching Assistant

The video demonstrates a prototype built using free, open-source LLM software, run entirely on a local computer [00:00]. This model is designed to work with documents created in a specific workspace dedicated to SEN and pupil data analysis [00:06].

For demonstration purposes, the system was fed a variety of fictional reports—including behaviour incidents, exclusion reports, and timeline events—for a pupil named ‘Bruce Banner’ [00:24].

How the Prototype Works: From Data to Insight

The process is straightforward:

1. Staff Query (The Prompt): A new member of staff enters a direct query, known as a ‘prompt,’ into the system [00:50]. Prompt writing is highlighted as the most important part of the process, as the quality of the question dictates the quality of the answer [00:57].
2. Local Processing: The LLM processes the prompt using the documents saved within the secure local workspace. Unlike cloud-based LLMs that use super-powerful computers at large multinational companies, this prototype runs on the local machine, resulting in a slightly slower, but far more secure, processing time [01:13].
3. Immediate Summary: The system returns a comprehensive summary of the pupil’s key data points and events, directly addressing the new staff member’s needs [01:29].

The Power of Transparency: Citing Source Documents

A key feature of this local LLM is its commitment to transparency. Every piece of information in the generated summary is accompanied by a citation that links back to the original source document, whether it’s a behaviour report or an exclusion file [01:21].

This is a vital safeguard in an educational setting:

• It builds trust in the information provided.
• It allows the staff member to easily verify the details.
• It ensures the tool is acting as an assistant to human judgment, not a replacement.

Conclusion and Open Conversation

This prototype offers a glimpse into the future of educational administration, where efficient data management and staff support can be achieved with secure, locally-run Large Language Models. By quickly providing essential context, schools can ensure new Teaching Assistants are ready to support vulnerable children more effectively from their very first day.

I am interested in opening a conversation on this topic and hearing from other schools who are exploring similar uses of LLMs in education [01:45].

---

About the Author

Enzo Vullo (Mr Vullo) is a qualified teacher currently serving as the Head of an SEMH provision in Harlow, Essex, United Kingdom. His provision supports vulnerable primary-aged children who find it challenging to access learning in mainstream settings.

With previous experience as a SENCo and a class teacher in both mainstream schools and a pupil referral unit (PRU), Enzo is passionate about using technology—specifically coding and STEM—to re-engage children in learning, develop resilience, and facilitate social mobility. He is committed to giving children the tools to succeed in education.