Capability layer: AI models
The requests your work receives are the one thing a general model was never trained on.
The model is one part of a system Sheba also runs, never a product on its own. Sheba builds models and fine-tunes them for the use case they will serve inside its products and the deployments it operates.
The work itself
The model is fine-tuned on data that represents the use case, then placed in the system it was prepared for. What that data has to represent is settled first: the requests the model will meet, and the language they arrive in.
Sometimes the work is building a model and sometimes fine-tuning one that already exists, which is why not every model here starts from scratch.
Languages and dialects
Part of fitting a model to a use case is fitting it to the language, and the dialect, its requests arrive in.
One example of this work
One market, and a model built for it
It runs inside Sheba’s systems rather than on its own, and it shows what this layer means in practice.
Where the models run
None of them is offered on its own, because each was fitted to work inside a particular system: a Sheba product, or a deployment Sheba builds and operates.
Choosing and routing between models
That belongs to Sheba Enterprise Platform, which is production-ready, and its own page describes it: which model takes which task, and how requests are routed between models.
See how the platform routes modelsFAQ
What people ask about the model work
Whether your use case needs a fine-tuned model is the first thing to find out.
Scoping the model work comes before any estimate, and it starts from the use case you describe.
Discuss model work for your use case