THEIR CHALLENGE
Streamlining access to a wealth of library content in record time
Swisscom is Switzerland’s leading provider of mobile, internet, and TV services. In a country of nearly nine million people, Swisscom has more than six mobile connections and two million broadband customers. It’s a household name, but it’s also so much more than a consumer-facing brand. Swisscom provides information and communications technology services to some of the country’s leading enterprises and the buoyant small and medium-sized enterprise sector. It sees its role as empowering Switzerland’s digital future and claims to be the most trusted Swiss technology innovator.
The business plans to invest CHF100 million ($117M) in AI solutions in the coming years. It is working with Nvidia to build generative AI (gen AI) full-stack supercomputers, based on Nvidia accelerated computing and Nvidia AI software. Swisscom wants to take a leadership position in AI, helping find and develop go-to-market AI services relevant to Swiss customers.
As part of this, Swisscom is keen to embrace AI within its own operations and is exploring several use cases.
One particular case is e.foresight, Swisscom’s independent “think tank for the banking industry.” The financial services sector accounts for nearly 10% of the Swiss gross domestic product; e.foresight identifies, monitors, and analyzes trends and behavioral and technological developments in the global financial sector. Used by 30 of the country’s leading banks, it has built up a sizable library of expert content over the years.
Swisscom is using MongoDB Atlas to apply gen AI to this extensive library. The engagement enables Swisscom to extract maximum value from its content, producing bespoke content for the hundreds of e.foresight users within the 30 banks subscribing to its services. By doing so, it is transforming the speed, accuracy, and relevancy of search requests. Rather than forwarding six reports for users to pore over, it can use gen AI to produce a condensed summary within seconds.
“e.foresight has over 3,500 documents about trends in the banking industry in its library, with more being added each day,” said Dr. Martin Lorenz, Product Manager, Banking Front Solutions, Swisscom. “MongoDB Atlas enables us to store and search all that information and create something relevant to a specific request.”
OUR SOLUTION
Enriching large language models with MongoDB Vector Search on Atlas
In practical terms, the Swisscom gen AI application needed to enable large language models (LLMs) to make sense of the e.foresight document library. The library is not only large but also diverse. It includes unstructured data, such as written reports, recordings of live events, graphics, and external studies, and it continues to grow.
MongoDB was critical to achieving Swisscom’s goal by helping power its retrieval-augmented generation (RAG) application. Now all documents in the e.foresight library are stored in MongoDB Atlas—a single, secure database for storing and searching vectors alongside operational data. Swisscom can then convert the unstructured data into vector embeddings and use MongoDB Vector Search on Atlas to find segments of relevant context to feed the LLMs up-to-date contextual, domain-specific data. As a result, the application can generate more accurate, highly relevant responses for users.
In a fast-moving industry, with constant breakthroughs in gen AI capabilities, MongoDB empowered Swisscom to move quickly from idea to production, enabling faster speed to market. Lorenz reached out to MongoDB on February 6, 2024, and had a prototype based on a subset of data fully built on the MongoDB Atlas Developer Data Platform by February 15. After a month of testing alongside MongoDB and the university of applied science ZHAW, the Swisscom internal team took over the platform operations.
The team added new features, and after another month of in-house testing, the platform went into production in early May.
.svg)