If you are considering applying for the Scottish public sector AI Challenge, it helps to understand the feasibility of your potential use cases by considering how it can be implemented from a technology perspective. This article provides an overview to what might be a suitable route to consider depending on your particular use case.

It is worth saying from the outset that there are a range of AI technology solutions available to public sector organisations and there is no one-size fits all approach.

For ideas that hopefully go on to be shortlisted at the Proof of Concept (PoC) stage of this year’s challenge, technical assessment – including model comparison and selection – will take place with assistance from Storm ID, as Futurescot’s challenge partner, and a range of cloud-hosted and open-source models will be considered based on the applicability and performance of each model to support the development of the use case.​

For the purposes of this article, we’ve chosen to outline three of the most viable routes which we believe are delivering the greatest value from AI in our work across public sector.

Option one: Microsoft Copilot

Microsoft Copilot is an AI assistant built into M365 ecosystem and with Copilot Studio you can build workflows to automate certain processes. It’s increasingly a good route for certain use cases given many public bodies already have a large amount of data stored within M365 or SharePoint and connectors to other data sources are also available.

Consider when: Your use case will rely heavily on data from SharePoint, Outlook or the wider M365 eco system.

Limitations: Depending on your workflow complexities or specific internal system integrations, it may not be suitable.

Option two: Custom AI Agents using Cloud AI models

In a previous article we outlined the potential value of AI agents for public sector usage and shared some examples of these. Adding increasing workflow complexity and specific internal business logic tends to merit a more bespoke approach to building AI workflows with whatever cloud-based large language model (LLM) or combination of LLMs are suitable for the task and are available in your cloud platform (e.g. Azure OpenAI).

Consider when: You have quite specific and complex workflows which you are seeking to automate and which may rely on multiple data sources. To help you, this guide outlines some suggestions on how to design a custom AI agent for your application.

Limitations: It’s likely this approach will require more development time, however for the purposes of the AI Challenge StormID have an accelerator environment where we can quickly prototype custom AI workflows using common pre-built components.

Option three: Private AI using open source models

For scenarios where data sensitivity, privacy concerns or specific regulations necessitates that data remains off public cloud providers, you can consider Sovereign or Private AI options.

This would involving deploying AI systems within your organisation’s own network, connected directly to your data using open-source models. If this is a suitable option, we can also demonstrate this in the PoC process for shortlisted organisations using the StormID AI accelerator.

Consider when: Data is too sensitive to be in a cloud platform or to interact with cloud AI models.

Limitations: Open source models currently lag behind the latest closed cloud models for general purpose use cases, although that may change in the future.

AI clinic

At StormID we have invested in developing a range of AI accelerators and subject matter expertise across a broad spectrum of AI, so we can support you in rapidly demonstrating the appropriate technology options via the PoC process. More information on the PoC process is outlined in this article.

If you would like to book a free 30-minute AI Clinic to review any use case ideas or get technical guidance simply email ai@stormid.com with an outline of what type of guidance you would like.

A meeting will typically cover:

  • Discuss your use case ideas or problems you are trying to solve with AI
  • Provide feedback around feasibility including technology / model selection support
  • Suggest any possible enhancements to improve the AI Challenge application

As a reminder, the deadline for applications for this year’s Challenge is August 22. For more information on the challenge click on the banner below. Good luck!