Designing Mojo
Prototype exploring how LLMs can support leadership teams
Client: Internal (Project Imagine)
Role: Product strategy, design and experience
Building a chat-to-your-data interface
In Early 2023, I proposed we should start exploring what we could do with LLMs as I knew our banking platform was pretty AI ready. From experience, we felt it would be hugely valuable leadership teams to have direct access to their data, so we set out to build an infrastructure that would allow us to 'chat to our data'. I worked with the front-end dev to setup a basic component schema using material flutter widgets. This would allow us to build at speed and focus on the technology.
Surfacing LLM ‘thoughts' to build trust
While working with the backend devs, we could see through Langchain that there was a lot more going on 'under the hood' that was initially being surfaced. Bringing these 'thoughts' and 'actions' to the front end, so the user can see what's happening, builds trust in the answer the LLM finally produces. (This was before ChatGPT was doing anything similar). It had an added benefit of reducing the perception of latency, that was quite an issue in models at the time.
Potential for a leadership companion
Much of our thinking was based on our own experience leading a twice-regulated fintech, but we also interviewed leaders in different sectors to look at how pervasive needs were.
• Barriers to data
• Loneliness – buck stops with them, few if any people they can turn to in confidence
• Lack of prep time for meetings
Were all common concerns. Because of these we added a private, General Chat feature, Doc Upload & Analysis feature, and Procedures – an early form of agent.
Outcomes of the ‘Mojo' prototype
The build for Mojo was recognised by a US startup, who brought our team in to work on Finley.
Mojo was one of the first GenAI interfaces that queried live production data via real-time database mapping and SQL generation. It broke down barriers to business data while demonstrating potential to support leadership teams.