Responsible AI at Tessavra
Tessavra is designed so that a person stays in charge and every finding points to the line behind it. This page says what exists in code today and what is still being built.
Human review
A finding can carry a review state that a person sets: not reviewed, in progress or reviewed. AI scoring of support calls is in development and not ready, so today a manager reviews a support call and records the result.
Evidence you can check
A finding is designed to cite the line that produced it. Where it cites audio, the citation carries a confidence marker such as Exact or Unverified, so nobody mistakes a guess for a quote. Citation accuracy has not yet been evaluated on a labelled set of conversations.
Guidance your team writes
Live guidance in Cortex comes from rules your team writes or passages retrieved from your own documents. A model does not compose the next thing a rep should say. Cortex is in development.
Features stay off until granted
A feature is off for an organisation until it has been granted, and a grant that cannot be resolved leaves it off. This has been tested on a local stack, not in a hosted environment.
Trying a rule before it reaches anyone
Cortex can run a rule against a past conversation without notifying anyone, so a team can review what the rule would have done before it is switched on.
- Write the rule
- Try it on a past conversation
- Review what it would have done with your team
What this page does not claim
Some features send audio or text to third-party speech and language-model providers, and no provider list is published. We make no promise about how those providers handle data, and we do not claim that conversations are never used to train models. No claim that findings are accurate, no approval step before an AI action, and no certification, including under any AI regulation.
Bring your questions on AI use.
Request a demo and we will show what exists today, what is in development and what is not built.
Request a demo