Capabilities
What the models actually do
Five bounded tasks that make up most of a regulated call. Each is small enough to run locally and specific enough to benchmark.
- Multilingual & script fidelityCode-switch handling · correct script per language run
- Objection handlingClassify the pushback · respond in approved language
- Compliance guardCheck before speaking · log the verdict
- Call structuringSchema-valid output into your CRM
- Voice latencyThe engineering that makes the other four fit in a turn
Each capability is trained per industry. The insurance objection model and the real estate objection model share architecture and nothing else.