When an incident occurs on a plane or at an airport, a report is written to document the event.
Daugherty’s partner, a major airline, saw an opportunity to be more proactive and use this data to help identify trends and better understand what incidents have occurred, where have they occurred, who or what was harmed, and why.
Safety and security incidents are wide-ranging and vary from situations like: inebriated passengers, equipment malfunction, too many passengers congregating at the front of the plane, back injuries while loading luggage, or a customer tripping on the strap of a handbag and breaking their arm. The information gathered is necessary to help resolve issues, settle disputes, and meet requirements for regulatory agencies like FAA, OSHA, and TSA.
Combining regex and GenAI tools, a team of Daugherty data scientists partnered with the airline to build an integrated system that allows users to search semantically similar documents and narratives for types of incidents and categorize the results for real-time tracking – replacing manual keyword searches with AI techniques to identify non-obvious correlations and causes.
The team reduced compute time by over 90% on a GPU instance, increased performance in returning narratives by over 20%, and mapped out a feedback loop which will provide input on model performance and enable improvement over time.
The increased computing efficiency from the GenAI tools proactively identifies trends and is helping the team develop and implement strategies aimed at keeping employees and passengers safe from harm.
Along with technical improvements, Daugherty’s team worked to protect confidentiality of passengers and employees by further developing tools for redacting personally identifiable information (PII), such as names and other sensitive data.