Trillet

Solution

Trillet moved its voice application layer to Google Cloud so conversational automation could be both predictable and scalable. Gemini in Vertex AI became the core reasoning engine, allowing the team to run with near-zero model freedom for high-stakes healthcare and enterprise interactions while still maintaining large context and fast responses. The models cross-reference live clinic databases and scheduling rules to formulate answers, which sharply reduced erroneous appointment responses compared with earlier LLM attempts. Google Kubernetes Engine and Cloud Load Balancing provide the elastic runtime needed to absorb large call spikes, while the unified Google Cloud stack removes the operational overhead of a fragmented black-box setup. This architecture lets Trillet orchestrate thousands of calls simultaneously, keep error rates below one percent, and combine real-time data access with strict AI control in a way that is suitable for regulated customer interactions.

Data flow

Caller requests enter through Cloud Load Balancing into services on GKE, where Gemini models on Vertex AI validate the request against live clinic data and scheduling rules. The resulting response is returned to the caller or escalated if needed, while the platform elastically scales for spikes in call volume.