Work

What an engagement looks like.

The three engagements below are written as examples. They are hypothetical. We do not publish client names, logos, or quotes.

audioSales

Scenario 01 / Sales

Discount approval calls

Question: When does a rep have the room to discount, and when do they need a manager?
Protocol: Call audio across 300 sales calls, four reps, marking the moment a discount was offered and who approved it.
Output: A decision trace per call showing what triggered the discount, with an eval on a blind set of 50 calls.

videoRetail

Scenario 02 / Retail

Return desk video corpus

Question: What makes an associate approve a return with no receipt?
Protocol: Ninety days of counter video across two stores, aligned to a spoken reason at every approved and denied return.
Output: A labeled video corpus with the associate's reason attached to each call, and an eval built from the returns that were later reversed.

audioSupport

Scenario 03 / Support

Escalation dataset

Question: When should a support case leave tier one?
Protocol: Call audio and ticket state across 1,200 cases, with the agent marking the moment they judged the case beyond their tier and saying why.
Output: An escalation dataset with a human decision point per case, and an eval scoring a model against the agents on held-out weeks.

Your turn

The fourth scenario is the one inside your company.

Backing

Backed by Y Combinator and General Catalyst.