Sotto
Client: UK hospitality operator (under NDA)
AI voice ordering for restaurants. Modular .NET service set, low-latency conversational pipeline, GDPR-clean audio handling.
Definition
What is the Sotto case study?
Sotto is a LoneSock-built AI voice ordering platform for restaurant operators. It handles phone and in-venue ordering through a natural-language conversation, routes orders into existing POS systems, and processes voice without persisting customer audio so the deployment stays GDPR-clean.
The Challenge
What we were solving for.
The operator wanted to absorb peak-hour phone volume and walk-in ordering load without growing staff headcount. Their existing IVR was inaccurate, slow, and consistently rated as the worst part of the customer experience.
Any voice solution had to behave conversationally rather than menu-tree IVR. It needed to handle hesitations, modifications, allergens, and accents in real conversation, and it needed to stay inside GDPR boundaries for voice data, with no audio leaving the processing region.
Integration was non-trivial. Different sites used different point-of-sale systems with different APIs, different menu models, and different timing assumptions for kitchen handoff.
What We Built
Technical architecture.
A conversational state machine that drives the full ordering lifecycle from greeting through confirmation and payment routing. Each transition is shaped by LLM inference inside a hard latency budget so the line never feels dead.
A modular .NET service set organized by domain boundary. Real-time audio uses WebRTC for in-venue interfaces and SIP for phone calls, with SignalR pushing kitchen-display updates as the order forms.
An adapter layer that fronts multiple POS systems through a single internal contract, so the same conversation pipeline can dispatch orders into Toast, Square, or a proprietary backend without changing the upstream flow.
A test harness that exercises voice recognition, conversation flow edge cases, POS contracts, and GDPR data-handling rules on every deployment gate.
Key Numbers
Attributed, honest figures.
We publish numbers that the engagement supports. Unverifiable marketing metrics are not on this page.
Technology Stack
What we built it with.
Outcome
What the engagement delivered.
Ordering conversations run with low-latency turn-taking, which keeps the interaction feeling natural rather than mechanical.
Audio is transcribed in memory and discarded. No recordings are persisted in the pipeline, which materially simplifies the GDPR posture.
The POS adapter layer means new sites can be added without rebuilding the conversational core.
Questions
FAQ for the Sotto engagement.
What did LoneSock build for Sotto?
LoneSock designed and built the full Sotto voice-ordering platform: the conversational state machine, the .NET service set, the WebRTC and SIP transport layer, the POS adapters, and the deployment and test harness. The operator owns the resulting system and the source code.
What tech stack was used?
.NET 10 services, a Groq-served LLM for conversation, SignalR for real-time UI updates, WebRTC for browser-based audio, SIP for phone-line ingress, and PostgreSQL for state. Deployment runs in Docker.
Is the system GDPR-compliant?
Yes. Voice data is transcribed in memory and discarded. No audio recordings are persisted anywhere in the pipeline, which keeps the personal-data footprint to transcripts only.
How long did the engagement take?
It is a multi-quarter engagement. The initial build shipped a working ordering loop, and LoneSock continues to iterate on accuracy, POS adapters, and menu coverage.
Can Sotto integrate with our existing POS?
Yes. The platform sits behind a POS adapter layer, so the same conversation pipeline can dispatch orders into different point-of-sale systems with site-specific configuration rather than a code rewrite.
Working on something similar
We have shipped this class of system before. Tell us what you are dealing with.
$20/hr time and materials, fixed-price milestones, or retainer. Source code is yours.