APIsolutely Awesome: Building a Gen AI Data Hub in Low-code

Growing up in the 80s, my first exposure to computers was loading ZX Spectrum games from cassette tapes or typing out BASIC code on the school’s single BBC terminal. There have been several defining leaps forward since then: the internet, mobile phones, and cloud computing. It somehow feels like artificial intelligence is going to be an even bigger change to our lives, however.

Generative AI (Gen AI), is revolutionising how we work too. At Adur & Worthing Councils, we’re constantly looking for ways to make the most of these advances to improve our services. It is true that it has taken us some time to fully embrace the potential of AI, but we are now very excited to be working on several proofs of concept.

It was important for us to get the building blocks right first, which is why we built a central hub (an API abstraction layer) using our low-code platform, Liberty Create. This hub will let us easily connect many of our existing applications across our various product environments to Google Gemini AI and other Gen AI tools in the future.

Why an Abstraction Layer?

As we explore the possibilities of Gen AI, we see the need for a flexible and scalable approach. Directly integrating individual AI APIs into our applications can lead to tight coupling. This means if we wanted to change one tool, we’d have to change lots of connected systems as well. It would make it difficult to switch or upgrade AI providers in the future. An abstraction layer acts as an intermediary. It provides a consistent way for our applications to interact with various AI services. This approach has several benefits:

• Flexibility: Easily switch between different AI providers without changing our core applications.
• Scalability: Handle more AI tasks and grow in the future.
• Maintainability: Simplify updates and make managing multiple AI connections less complex.
• Consistency: Ensure a uniform experience for our users across different AI-powered features.

Our Initial Focus: Google Gemini AI

Our first step is to use Google Gemini AI, a powerful new Generative AI tool. We’ve initially chosen Gemini due to our confidence around data security, with data staying within our existing Workspaces platform and not being used to train new AI models. However, as different AI platforms improve in specific areas, we can easily adapt and onboard new API connections.

Diagram: Simplified Architecture of our Product Environments connecting to AI Platforms

Example internal request to Control:

{
   "prompt_id":1234,
   "app_directory":"Report it",
   "service":"Waste",
   "model":"gemini-1.5-flash",
   "description":"Ward reports summary",
   "request_contents":"Summarise this....",
   "request_systeminstruction":"",
   "request_generationconfig":""
}

Example internal response from Control:

{
   "prompt_id":1234,
   "response_text":"This ward has....",
   "response_request_tokens":50,
   "response_response_tokens":25,
   "response_total_tokens":75,
   "response_block_reason":""
}

Building the Central Hub on Liberty Create

Our low-code development platform is perfect for building our central hub. We can quickly develop and deploy the necessary components, including:

• API endpoints to receive requests from our applications.
• Logic to route requests to the correct AI service.
• Data transformation and mapping to ensure compatibility between our applications and the AI tools.
• Error handling and logging to monitor and troubleshoot issues.

Trialling some Examples

We’ve identified several possible use cases where Gemini could significantly improve how we work. These are connected to the following products:

• Report It: From the data in our product for reporting issues in a neighbourhood, we will trial the production of summary reports. These could summarise issues by Ward level, highlighting key themes such as the amount of graffiti or fly tipping.
• Citizen Hub: From our customer and contact centre platform, we will test transcribing and summarise call recordings, saving time on manual summaries and improving agent training.
• Complaints: Categorise and summarise complaints to identify trends and areas for improvement.
• Requests for Information (RFI): Suggest Senior Information Officers (SIOs) for RFIs based on their expertise and team responsibilities.
• Housing Voids: Draw attention to where action is needed and generate project summaries, paying particular attention to overdue or target dates.
• Housing Repairs: Summarise repair projects to provide quick overviews, progress and insights for officers.

Expanding to Other Gen AI APIs

While we’re starting with Google Gemini AI, our hub is designed to be expanded. In the future, we plan to connect with other Gen AI tools as needed, giving us more flexibility and choice. This could include:

• Exploring AI models for specific tasks, such as image or video analysis.
• Using AI services from different cloud providers.
• Adapting to new and emerging AI technologies.

Conclusion

Building a central hub for connecting our APIs to AI on Liberty Create is a strategic choice, similar to how we previously used it with GOV.UK Notify for sending SMS messages. By starting with Google Gemini AI and planning for the future, we are creating a flexible, scalable, and maintainable foundation for AI-driven innovation at Adur & Worthing Councils.