Sotto
AI voice ordering for UK restaurants with sub-500ms response times
The Challenge
What we were solving.
A UK restaurant group needed to handle high volumes of phone and in-venue orders without increasing staff. Existing IVR solutions were slow, inaccurate, and universally hated by customers.
The system had to understand natural speech patterns -- including accents, hesitations, and menu modifications -- while remaining GDPR-compliant for all voice data processing. No customer audio could be stored or transmitted to non-compliant jurisdictions.
Integration with multiple POS systems (Toast, Square, proprietary) was required, each with different APIs and data models. The ordering flow had to feel conversational, not robotic.
What We Built
Technical architecture.
We engineered an 8-state conversation machine that handles the full ordering lifecycle: greeting, menu navigation, item customization, upselling, order confirmation, payment routing, and post-order feedback. Each state transitions are driven by Groq LLM inference with hard latency budgets.
The architecture spans 29 .NET projects organized as a modular monolith with clear domain boundaries. Real-time audio streaming uses WebRTC for browser-based ordering and SIP trunking for phone calls, with SignalR managing bidirectional state updates to the kitchen display system.
We built a comprehensive test harness with 649 automated tests covering voice recognition accuracy, conversation flow edge cases, POS integration contracts, and GDPR data handling. The test suite runs in under 3 minutes and gates every deployment.
Key Numbers
By the numbers.
Technology Stack
Built with.
Outcome
Business results.
The system processes voice orders with sub-500ms end-to-end latency, making conversations feel natural and responsive. Order accuracy exceeds 96% even with complex modifications.
Full GDPR compliance achieved with ephemeral audio processing -- voice data is transcribed in-memory and immediately discarded. No audio recordings are persisted anywhere in the pipeline.
Multi-POS integration allows the restaurant group to use Sotto across all locations regardless of their existing point-of-sale infrastructure.
Working on something similar?
We've solved this class of problem before. Let's discuss your specific requirements.