Seldon.
Seldon is a model deployment platform for Kubernetes supporting inference graphs — chains of models, transformers, routers and combiners — as well as explainability and outlier detection components.
Where Seldon fits on a long engagement.
Inference graphs are Seldon's distinguishing capability. Real systems are rarely one model: a preprocessing step, an ensemble, a router choosing a model by segment, and a combiner merging outputs. Expressing that as a deployed graph is more honest than hiding it in application code.
Explainability and outlier detection as first-class components matter in regulated contexts. A prediction served alongside an explanation, and inputs flagged as outside the training distribution, are requirements in several sectors rather than refinements.
What an assigned team does with Seldon.
Graph complexity carries operational cost. Each component is a service to deploy, monitor and version, and a graph with six components has six things that can fail and a combinatorial set of version interactions.
Keeping graphs as simple as the problem permits is a design discipline held by the assigned specialists, as described in how an assignment runs.
What we use Seldon for.
- Multi-model systems deployed honestly Routing and ensembling expressed as a graph rather than hidden in application code.
- Explanations served with predictions Explainability as a component, where regulation requires it.
- Outliers flagged at inference Inputs outside the training distribution detected rather than scored confidently.
How Seldon capacity is assigned.
Serving architecture is assigned inside AI capacity, with graph complexity kept proportionate to the problem.
Tell us what your roadmap needs Seldon for.
A service delivery manager replies with the disciplines we would assign, the monthly capacity and what the first month looks like.
Loading the contact form… You can also email hello@azendo.co.
We reply within one working day. No obligation, and no newsletter.