AI Challenge with SBB: Working Together from the Idea to the Implementation

In the summer, SBB invited us to take part in a challenge for companies specializing in artificial intelligence (AI). We’ve been working with SBB for years. Throughout this time, we have focused on challenging use cases, highly complex problems, and genuinely relevant solutions – in other words, projects with opportunities to harness AI’s full potential. The AI Challenge was a chance to develop forward-looking solutions and translate innovation into tangible ideas. Most importantly, it would allow us to share our expertise in implementing powerful AI applications effectively. In short, we absolutely wanted to be involved.

The aim of the challenge was to support employees at the SBB Train Control Center with AI-based solutions.

The Train Control Center is where all SBB’s rail operations come together: Its employees monitor operations, respond to disruptions, and coordinate trains, engine drivers, and other operators. They make sure rail traffic keeps running even in exceptional situations. To do this, they often have to make decisions under time pressure and based on multiple streams of information.
 

For this challenge, we consciously opted for a compact, collaborative format: a small hackathon consisting of three AI Nights. Over the three evenings, around 20 highly motivated ti&m employees worked together on tangible solutions. To assist them, they had support from experts from SBB’s Train Control Center. The first AI Night took place at the SBB Train Control Center in Olten. The second and third sessions were held at our Bern office.

AI Night #1: Understanding the challenges and finding solutions
The first evening in Olten was all about understanding. SBB employees explained what happens at the Train Control Center. We were also able to observe employees in their daily work. Together, we identified the main challenges in everyday operations and discussed areas where AI could provide tangible support.
Based on this, we came up with various possible solutions. We then assessed their benefits, feasibility, and relevance with SBB.

AI Nights #2 and #3: Implementing the solutions in interdisciplinary teams 
At the second AI Night, we moved on to implementation. Working in teams, each group focused on a chosen solution and implemented it as a prototype. On the third AI Night, we further developed, refined, and consolidated the results.

The challenge produced three solutions at different stages of development:

A working solution
This solution consists of a transcription tool that combines a speech recognition model with a large language model (LLM). The solution automatically transcribes verbal communication between Train Control Center employees and train drivers, for example. The tool then prepares the content in a structured manner. This means that information can be recorded, tracked, and processed faster.

A solution with potential for development
After the AI Nights, we developed another solution further: an AI-based decision-making aid. We produced a proof of concept that demonstrates its potential. To create this prototype, we trained a model based on historical data and previous decisions. In the event of disruptions, this model recommends actions for specific trains. This gives employees the benefit of experience when making decisions, especially in time-critical situations.

A concept for major incidents
Our third solution is a conceptual approach. It is especially relevant to major incidents, such as large-scale disruptions. In such situations, we showed how automated recommendations can be made based on existing SBB solution concepts. We think this approach has a lot of potential. However, we decided not to invest in it further. In this case, working closely with SBB on a dedicated project would be the best way to implement the concept.
 

Finally, we presented our solutions to various SBB stakeholders and other participating companies at a plenary session in mid-December. Together, we discussed the solutions and considered SBB’s feedback. The event was a chance to look back on a successful AI Challenge – and raise a toast to the results achieved.