Grokking LLM Evals
Vote

0% completed

​

Trajectory Checks

Every run of the scheduling agent must keep five promises about its path:

  • It never calls create_booking a second time for a ticket that already has a booking.
  • It checks that a technician is free at a time before it books that technician at that time.
  • On an appliance ticket that names a part, it checks parts before it books.
  • It finishes within 15 tool calls.
  • Every tool call uses valid arguments: one of the six trades, one of the 12 technicians, and a real day.

It also makes three non-promises

.....

.....

.....

Like the course? Get enrolled and start learning!

Reading Progress

0%

Vote for new content