Caylent Catalysts™
Generative AI Strategy
Accelerate your generative AI initiatives with ideation sessions for use case prioritization, foundation model selection, and an assessment of your data landscape and organizational readiness.
United Airlines Improves Flight Operations with Generative AI on AWS
Generative AI
At a glance
United Airlines, one of the world's largest airlines operating approximately 4,000 flights each day, partnered with Caylent to transform how its engineering teams monitor and analyze operational data for its mission-critical Weight & Balance (W&B) system. By building an operational intelligence platform on AWS, United reduced mean time to detect and resolve issues by approximately 90%, cut investigation times from 10 hours to 1 hour, and returned hundreds of engineering hours each month to higher-value work.
Solution Implemented
→ Applied the AWS Business Value Realization framework to identify the highest-impact AI use case.
→ Used Amazon Kinesis to stream operational log events and convert them into searchable vector embeddings with Amazon Titan on Amazon Bedrock, indexed in Amazon OpenSearch.
→ Built a web dashboard that allows engineers and business users to ask questions in natural language, using a two-stage search (metadata filter, then semantic search) and Anthropic’s Claude models on Amazon Bedrock to generate answers.
→ Built integrations into United’s internal enterprise AI gateway to meet enterprise security and governance requirements
Outcomes
→ Reduced mean time to detect and resolve operational issues by approximately 90%.
→ Reduced investigation time from roughly 10 hours to 1-2 hours.
→ Eliminated approximately 336 hours of manual log analysis each month.
→ Enabled engineering teams to quickly analyze approximately 30 million operational log events generated daily.
→ Improved developer satisfaction by allowing engineers to focus on innovation rather than manual troubleshooting.
Turning Operational Data into Actionable Intelligence
Every day, United Airlines operates approximately 4,000 flights across one of the world's largest airline networks. Behind every departure is the airline's Weight & Balance system, a mission-critical application that ensures every aircraft is properly loaded before takeoff. Because every departure depends on accurate Weight & Balance calculations, the system generates an enormous amount of operational data: approximately 30 million log events each day.
While this wealth of telemetry contained valuable operational insight, extracting that information proved increasingly difficult. Engineers were spending significant amounts of time manually searching logs across multiple systems to investigate anomalies, identify root causes, and understand emerging operational trends. The process was reactive, time-consuming, and diverted engineering resources away from delivering new capabilities.
United recognized that solving this challenge required more than better monitoring. The airline wanted to transform operational data into an intelligent system capable of identifying issues proactively, accelerating investigations, and continuously improving operational performance.
Business Challenge
United's engineering teams supported a highly distributed operational environment where millions of log events flowed through multiple applications and external systems every day. Although the airline had extensive monitoring capabilities, identifying meaningful issues still required engineers to manually sift through massive volumes of operational data.
This manual process consumed approximately 336 engineering hours each month and often meant problems were discovered only after they had already impacted operations. Investigation cycles routinely stretched to nearly 10 hours, slowing resolution and increasing the risk of downstream flight disruptions.
As United continued investing in digital transformation, the airline needed a smarter approach that could automatically detect anomalies, surface meaningful insights, and give engineers immediate access to the information they needed.
“Because the airline industry is so complex, having the right partners like Caylent and using AWS services helps us propel into the future without running into any issues and be proactive and get ahead of that curve. So wherever those issues do arise, we’re ready to tackle them without any problems.”
Milan Thakker
Project Manager
Solution
Working together, United Airlines and Caylent applied the AWS Business Value Realization framework to identify the highest-impact opportunity for generative AI. The assessment identified operational intelligence for the Weight & Balance system as the ideal starting point, delivering immediate business value while creating a foundation for broader AI adoption.
Caylent built a generative AI platform using Amazon Bedrock that enables technical and business teams to ask operational questions in natural language and receive contextual answers in near real time. Amazon Kinesis streamed operational log events and converted them into searchable vector embeddings with Amazon Titan on Bedrock, indexed in Amazon OpenSearch.
When a user submits a question through the web dashboard, the platform uses a two-stage retrieval approach: first filtering operational data using metadata such as flight, date, or other relevant attributes, then applying semantic search to identify the most relevant information. Anthropic's Claude models on Amazon Bedrock use the retrieved context to generate a clear, natural-language answer. The platform also integrates with United's internal enterprise AI gateway, providing the governance and security controls required to operate generative AI within the enterprise environment.
Previously, teams had to search across millions of log records across multiple systems to piece together relevant data. Now, users can ask questions such as how many passengers and bags departed on a specific flight and receive an augmented response within seconds. The AI platform analyzes millions of operational logs and pulls together data from multiple sources to provide a complete operational picture, helping teams resolve questions faster, spend less time manually investigating data, and focus more on building features and driving innovation.
Rather than replacing existing monitoring tools, the solution augments them with generative AI, making operational data significantly easier to explore and enabling teams to uncover patterns, identify root causes, and resolve incidents more efficiently.
Results
By combining intelligent observability with generative AI, United Airlines fundamentally changed how its engineering teams monitor and respond to operational issues. Operational issues are identified and resolved more quickly, while engineering time previously spent manually analyzing logs is now focused on higher-value initiatives.
The platform reduced mean time to detect and resolve issues by approximately 90%, while investigation time fell from roughly 10 hours to just 1-2 hours. Engineers also eliminated an estimated 336 hours of manual log analysis each month, freeing valuable time for innovation and higher-value engineering work.
Beyond operational efficiency, the solution improved the daily experience for United's engineering teams. Instead of manually combing through approximately 30 million log events generated each day, engineers can use natural language to quickly investigate incidents, identify root causes, and uncover emerging patterns.
"The quicker we can identify issues, the quicker we can fix or prevent them to deliver excellent customer service in the moments that matter most," said Thomas Jacobs, Product Owner at United Airlines. "By implementing AI in our solutions, we've improved our operational efficiency and customer experience."
As United continues investing in AI-driven operations, the platform provides a scalable foundation for applying generative AI across additional operational workflows, helping the airline stay ahead of issues and continuously improve both employee productivity and customer experience.
What’s Next
United and Caylent are continuing to explore opportunities to expand the platform from reactive investigation to proactive operational intelligence. The next phase will focus on using AI to identify potential issues earlier by automating anomaly detection, alerting, and pattern recognition across operational data before they affect flight operations.
Technologies Used
Caylent's solution leverages Amazon Kinesis to stream operational log data, Amazon Titan on Amazon Bedrock to generate vector embeddings, and Amazon OpenSearch Service to support fast, contextual access to operational information. Amazon Bedrock and Anthropic's Claude models on Amazon Bedrock power generative AI inference and natural-language responses, while Amazon ECS hosts application services and AWS Lambda supports event-driven processing. The platform also integrates with United's internal enterprise AI gateway to support enterprise security and governance. Together, these technologies enable near real-time data ingestion, contextual retrieval, natural-language investigation, and faster issue resolution across United's mission-critical Weight & Balance system.
