VAmoS Energy: Testing Voice AI Agents in Real World Conditions

Voice AI agents in the production environment have difficulty beyond simple laboratory tests. Multiple customer requests need to deal with background voice interventions and users who run their account. The researchers developed a new benchmark to measure these realistic scenarios: VAmoS Energy.

• 100 realistic call scenario - There are 2–4 separate requests in each call on billing and assistance issues
• 16 vehicle support - Status management with Stripe payment system and Apache Fineract credit motor u -Efficiency verification challenges - Account access but opening after successful customer verification
• Real world noise - Background speech and interruptions test the performance of the agent

Benchmark is not only able to answer and answer a question of the agent, but also measure whether it can provide a perfect customer experience. During the billing query, a customer wants to change the payment plan is included in scenarios such as having to restart in authentication or repeating the request due to background noise.

This research aims to develop voice agents ready to use for production environment. VAmoS Energy will help engineers improve their systems with realistic difficulty degree and provide more reliable customer service.