WORK / 01MINI MODELS / SLMS
For narrow tasks
that run again and again.
Mini models
for specific work.
FOCUSED EXAMPLESCase ACase BCase C
→MINI MODELRun one task well
→SPECIFIC OUTPUTClassifyExtractRoute
We train smaller models for a narrow job so they can run at lower cost, with less waiting, and be measured on the cases that actually matter.
Talk through a use case ↗WHY A MINI MODEL?
Smaller by design.
Specific on purpose.
A mini model is not a generic assistant with a company name on it. It is scoped around one task, its representative data and its quality bar. That gives a team three things to optimize deliberately: cost per run, response time and accuracy on the task itself.
GOOD FITS
REPEATABLE / NARROW / MEASURABLEExtract structured information
↗Route work to the right team
↗Review documents against a defined standard
↗HOW WE WORK
Scope one job
Define the input, output, edge cases and the real quality standard for the task.
Train and compare
Test the mini model against representative examples, cost and response-time requirements.
Deploy with a measure
Put it into the workflow with a clear way to monitor task quality and improve it.
THE PRACTICAL TEST
If the task is not
clear enough to
measure, stop.
We design the evaluation alongside the model. The question is simple: does this model make this specific task cheaper, faster or more accurate for the people responsible for it?
Looking for workflow automation instead? ↗Have a narrow task worth training for?
Let’s look at the task. ↗