Partnering with Google Cloud partner Telana, Rentokil Initial developed "On The Go" (OTG), a generative AI-powered assistant designed for hands-free interaction. Using Gemini’s advanced capabilities and Telana’s engineering expertise, the team built a solution that utilises internal APIs to provide conversational access to customer data. A standout feature was the implementation of Gemini Live, which enables bidirectional streaming so technicians can talk to the app as they drive to their next site.

"The audio experience was really important," says James Tatham, Account Director at Telana. "The use of Gemini Live allowed us to have a super interactive channel with the technicians, which was a game changer in terms of how they interacted with the app." 

The development process was remarkably agile, moving from inception to a pilot in Buffalo, New York, in just six months. To ensure the tool met real-world needs, the team conducted "ride-alongs," shadowing technicians in the field to gather feedback and fix issues in real-time. This approach was critical in moving beyond consumer-level AI to an enterprise-grade tool that could operate in demanding environments—even in the middle of a New York winter.

Telana helped Rentokil Initial establish a flexible architecture with Firebase Hosting, Firebase Authentication, and Firebase Functions creating a fast and secure environment that powers the frontend. The ADK agent is deployed to Gemini Enterprise Agent Platform to help easily monitor performance, manage sessions, and scale as necessary. 

"We chose ADK for its ability to leverage multi-agent systems to orchestrate complex workflows," says Andy Whelan, AI Architect at Rentokil Initial. "We can now easily extend the platform’s functionalities as new features or data sources need to be integrated as independent agents without disrupting the core architecture."

Bidirectional streaming communications are connected through a regional Cloud Run service, providing the lowest possible latency and most flexible scaling capabilities. Finally, the API-first architecture helped bridge data silos to pull service history and recommendations into a single stream.