Quicktext runs its hotel AI on swappable models, including ones trained on its own guest conversations

Quicktext has processed nearly a billion tokens on aixplain. It builds AI that answers hotel guests: questions about rooms, amenities, services and the area around, at any hour. Since 2024, the models behind those answers have been swappable. Quicktext has moved from general models to models trained on its own guest conversations, and on to newer ones, without rebuilding its product.

The challenge: the best model keeps changing

Every guest conversation carries one hotel's information: its rooms, its services, its policies. The answer has to be right for that property, every time.

At hotel scale, the model behind those answers has to be accurate, fast and affordable on every message. And the best choice changes as new models arrive. A product tied to one model falls behind, or pays for a rebuild every time it moves.

Quicktext also had something no general model has: its own record of real guest conversations. It wanted a model trained on them, without becoming a model infrastructure company.

Why aixplain: the model is a setting

• Swappable models. Open or closed models from any provider, or models Quicktext owns, all behind the same interface. Changing the model is a change of configuration, so the product stays as it is.
• Fine-tuning with serving included. aixplain trains models on Quicktext's data and serves them, so a model trained on its conversations can replace a general one the same way.
• Engineers alongside. aixplain's team supported Quicktext's integration through the first year of production.

The solution: swap the model, keep the product

Start general. Quicktext started on a general-purpose model, with an open model running alongside it.

Train on its own conversations. aixplain fine-tuned open Mistral models on Quicktext's guest conversations, round after round, and fine-tuned a GPT model too. Each version was compared with the last before it took over.

Swap when something better arrives. As newer models came out, Quicktext tried them and swapped them in. Its product never had to change to use them.

Room to grow. On the same platform, Quicktext has explored new building blocks, such as a voice pipeline that goes from speech to a model and back to speech, and reading documents with OCR.

What's next: agents with swappable models

The same idea carries to agents. On the aixplain OS, an agent's model is a setting: switch it to a better, faster or cheaper model, or to one trained on your own data, and the agent keeps its owner, its rules and its history, with no rebuild. Inspector agents check every run, and every run is on record.

See aixplain at work

aixplain is the operating system for accountable AI agents you own. Organizations run more of their operations and services on agents, on any model and any server. Every run is checked against policy, and every agent answers to an owner.

Book a demo to see how your product and your agents can run on the model that fits today, and the next one.