Amadeus is a major software player in the world of travel. The company offers technology solutions to industry organizations, such as airlines, airports, and hotels, enabling them to manage plane ticket sales, accommodation offers, and car rentals, among other offerings. Essentially serving as the booking platform for these organizations, it serves 170 airlines and helps support more than two billion passengers each year. This represents a huge amount of digital traffic. Amadeus records one million transactions per second and employs 18,000 people of 150 different nationalities. Additionally, it saw a turnover of more than €5 billion in 2024. Given this scale, it’s imperative for Amadeus to detect and resolve any platform issues as soon as possible.
Smarter internal tools
To detect failures, Amadeus’s IT department of more than 2,000 developers relies on an internal tool called PIM (Problem tracking record, Investigation, iMproved). IT maintenance teams are the primary users of this investigation software, which aims to simplify another of the company’s internal tools: ALF (Amadeus Logging Facilities). These tools enable the company to monitor logs—files containing metadata that can be used to contextualize specific events.
Logs are a historical record of everything that happens within a system, including events like transactions, errors, and intrusions. And there are many of these logs within the Amadeus infrastructure. “The problem is that over the years, we have accumulated a mountain of information,” said Jérémy Junac, Principal Engineer at Amadeus. “And we thought that it was surely possible to exploit it in an intelligent way.” After all, the objective for developers is to focus on value-added tasks and not on detecting flaws, bugs, or problems, a time-intensive task that involves repeatedly performing the same checks and repairs.
IT incidents tend to repeat in any environment. In fact, if they’ve appeared before, they likely have been solved. This is why the team at Amadeus considered the possibility of storing the history of all encountered problems and their solutions in a database that could guide developers. “We had the idea of taking PIM to the next level, making it more intelligent and creating an iPIM, or Intelligent PIM,” said Junac.
MongoDB Atlas for text and vector search
Amadeus’s goal is to use iPIM to retrieve as much information as possible from all the logs and gather it in one database, where users can match any new problem against all available history, especially in the case of bugs that may impact multiple applications. To enable fast and easy searches, Amadeus chose MongoDB Atlas and two of its key features, MongoDB Atlas Search (providing full-text search capabilities) and MongoDB Atlas Vector Search, all in a cloud environment.
“When we encounter a problem, we open a ticket with the description of the bug,” explained Ayoub Imami, Software Development Engineer at Amadeus. iPIM retrieves this description, vectorizes it using AI, and records it in MongoDB. It’s important to note that vector data are numerical representations of complex data, such as text. iPIM uses OpenAI to convert this complex information into mathematical vectors to enable AI algorithms to process it more efficiently. Additionally, iPIM retrieves logs and stores them in MongoDB, where they serve as the source of truth.
Now the company’s MongoDB database contains textual and vector descriptions of all its historical problems. With this information, developers can intelligently use MongoDB’s full-text search feature and, in a mass of unstructured data, detect the descriptions of new bugs and compare them to those already present in the database. However, this was only the first step.
With MongoDB Atlas Vector Search, it’s possible to go even further. The description of a problem made by a developer in one language may be different in another. “In the context of a multilingual system, it quickly becomes ‘hell,’” said Junac. MongoDB Atlas Vector Search enables teams to query data based on the semantics or meaning of the data rather than the data itself. This is possible through the digital representation of any data in the form of vectors, which can then be compared to each other by means of sophisticated algorithms. With vectorization, linguistic problems disappear at Amadeus. The system frees itself from exact linguistic terms and focuses on their meaning.
Conclusion
Thanks to the MongoDB environment, Amadeus maintenance teams can resolve issues in minutes, helping propel the company forward and maintain its leading position in the competitive travel sector.
The new infrastructure at Amadeus, as well as the essential contribution of OpenAI, enables the company to sustain and secure its IT architecture and prepare for any new services that travelers and tour operators will no doubt require in the future.
To find out more, go to MongoDB Atlas Search.
_Spot.svg)
.svg)