Inference
Inference is the step where a trained model actually produces an answer for a new input.
What it tells you about a developer
Inference work means the developer has dealt with the running cost and speed of an AI product. That skill matters to any company whose model bill scales with usage, and it overlaps with ordinary performance engineering.
Inference is named for the model's action of inferring, or working out, an answer from a new input, based on patterns it learned during training. Training happens once, in advance, and is expensive and slow. Inference happens every time someone uses the model, and the cost and speed of inference is a real, ongoing expense for any company running an AI product. A developer working on inference is usually trying to make responses faster or cheaper.