What challenges did Trillet AI face before AI implementation?

Early voice assistants had high latency and spoke over callers. The system violated hard business constraints and hallucinated appointment bookings. Manual auditing of thousands of call logs created operational burden. The business was operating with about a 5% error rate and high infrastructure costs.

What was the impact of the AI solution?

Reduced error rate from about 5% to well below 1%. Cut infrastructure and operations costs by about 80%. Achieved sub-two-second latency. Resolved 85% of customer service calls without human intervention. Increased conversion rates by 10% for legal aid clients.

Source

GCP

Trillet AI: Gemini in Vertex AI + GKE to reduce call-center voice agent errors and costs

Use case typeCustomer service automationUpdated Jun 13, 2026

Trillet AI builds a voice application layer for enterprise that automates high-stakes customer interactions such as appointment rescheduling and government follow-ups. The company moved its infrastructure to Google Cloud, using Gemini in Vertex AI as the primary reasoning engine and Google Kubernetes Engine for autoscaling real-time voice traffic. It also used Speech-to-Text and Text-to-Speech for voice interactions, with the model ingesting extensive client-specific business logic and negative constraints.

Organization: Trillet AI
Industry: Tech & Comms
Location: United States
Published: May 2026

Reported outcomes

85%

calls resolved without human interventionAutomation & deflection

5 percentage points to under 1%error rate reduction80%infrastructure cost reduction2 secondslatency+10%conversion rate increase

Strategic outcomes

Risk & complianceReduced voice agent errorsCost efficiencyLowered infrastructure and operations costsSpeed & agilityAchieved sub-two-second latencyScale & capacityResolved most calls without humans

Primary read

Use case focus

Showing 3 of 3

1Conversational AI
2Customer Service Automation
3Voice AI

Early voice assistants had high latency and spoke over callers.
The system violated hard business constraints and hallucinated appointment bookings.
Manual auditing of thousands of call logs created operational burden.
The business was operating with about a 5% error rate and high infrastructure costs.

Migrated the platform to Google Cloud to improve control and latency.
Used Gemini in Vertex AI as the main reasoning engine with a large context window for client-specific business rules.
Deployed the core real-time application on Google Kubernetes Engine to scale automatically during traffic spikes.
Used Speech-to-Text API and Text-to-Speech API to support conversational voice interactions.

Technologies

Gemini in Vertex AI Vertex AI Google Kubernetes Engine Speech-to-Text AI Text-to-Speech AI Cloud Load Balancing

Reduced error rate from about 5% to well below 1%.
Cut infrastructure and operations costs by about 80%.
Achieved sub-two-second latency.
Resolved 85% of customer service calls without human intervention.
Increased conversion rates by 10% for legal aid clients.

Architecture

Trillet AI runs its real-time voice application on Google Kubernetes Engine and uses Gemini in Vertex AI as the primary reasoning engine. The model receives large amounts of client-specific business logic and 'negative constraints' to avoid invalid actions, while Speech-to-Text and Text-to-Speech services support live conversations.

Sources & evidence1

Groundedness: 5/5Type: Customer StoryPublished: May 31, 2026Publisher: Google CloudEvidence: PrimaryConfidence: High

Primary source

AI-generated summary. Verify important details with the linked sources before relying on this case.

Explore related AI use cases

Customer evidence

Trillet AI: Gemini in Vertex AI + GKE to reduce call-center voice agent errors and costs

Use case focus

Challenges

Solution

Impact

Similar cases