Agent suite for C3 Readiness
Background

C3 Readiness is responsible for aircraft predictive maintenance, but because maintenance must be done on the ground, all data must be processed and triaged during the aircraft's downtime.

Agent documenting raw sensor data for case generation
Agent sorting alerts into existing or new cases
All information gathered for maintenance engineers to fully understand the problem and solutions to historical, related cases
Approach

Biggest problem was balancing how much automation provides the most value and comfort for our users.

The consideration for each agent included: trigger event, the level of automation, and most importantly the error and fallback states for when something goes wrong.

Most decisions leveraged on how much information the agent manipulates and how much time it saved our users.

All agents must have an appropriate recovery path for their failure points
Example of recovery actions for agentic mistakes
Impact

Readiness agentic pipeline spans multiple fleets including: B-1B, C-5, KC-135, C-130, and C-17 aircraft.

On B-1B fleet alone, it 79 cases per year, avoiding 3,350 hours of maintenance total (42.4 maintenance hours per case), and translating to $8.8 million in annual value.