L’Équipe and Devoteam Use Generative AI to Create Olympics Podcasts on AWS

French media outlet L’Équipe faced a tight deadline of 30 days to deliver a series of 19 podcast episodes to cover the Paris Olympics. With its staff already occupied with covering the Games on a daily basis, the company’s editorial team turned to AWS Partner Devoteam and Amazon Web Services (AWS) to find a solution. Using Amazon Bedrock, Devoteam created an innovative solution that used L’Équipe’s existing web content to create audio content. Together, L’Équipe and Devoteam launched the company’s first podcasts to be created with generative AI (gen AI) on time and to the high-quality standards the company wanted.

With just a month to create and launch a podcast series focused on the Paris Olympics, L’Équipe found itself without the staff resources to deliver the work. This was an important project for L’Équipe. With the Olympics taking place in its home country—the first time in 100 years—it had a huge opportunity to generate interest and attract a passionate audience. Knowing that it needed both technical support and external expertise, L’Équipe turned to AWS Partner Devoteam to help it make and deliver the series.

AWS suggested that L’Équipe create the podcasts using gen AI because L’Équipe’s team didn’t have the skills or cloud competency to tackle the project. The partnership with Devoteam brought both the innovative thinking and technical skills that this project demanded. “L’Équipe didn’t have time to focus on the innovation with the limited time they had to successfully finish the project, so we had to collaborate and pivot to find a suitable way to move forward,” says Armel Negret, sports community lead for France at AWS. 

Despite the obstacles, L’Équipe and Devoteam were committed to finding a way to create the podcast series while matching the quality of human-delivered content. One key decision was that the AI would not generate any original content. To create the podcast, the team would re-use existing content from the L’Équipe website, created by the journalists. This content was then fed into the gen AI tool using Amazon Bedrock—the easiest way to build and scale generative AI applications with foundation models.

After the initial solution was built, Devoteam created some demo versions of what the podcast could sound like and shared them with the key stakeholders. The team had no room for error, because any mistakes or mispronunciations could raise issues of credibility and trust with listeners. The results satisfied the editorial team, which was impressed with the accuracy of the AI-generated spoken voice. Using the existing content was a significant enabler to building trust in the solution, because L’Équipe’s journalists were reassured that the gen AI tool was not creating brand-new content itself and were satisfied that there would not be mistakes or unverified information entering the scripts through AI hallucinations.

Devoteam quickly built a solution that was able to create these podcasts in a repeatable, scalable way. It needed to be a simple, user-friendly process, because there was limited time to upskill the L’Équipe team to be able to generate the rest of the podcasts themselves. It used Amazon Simple Storage Service (Amazon S3) to store the text summaries generated by Claude 3.5, and then AWS Lambda was automatically triggered whenever a file was added to or modified in Amazon S3. It automated tasks such as validating text summaries and converting text into audio. Finally, it used AWS Step Functions to orchestrate the whole process, step by step, to ensure a smooth and structured production process for the podcast. 

After it had a prototype in place, Devoteam could easily show L’Équipe how to repurpose and input the journalists’ content from the website into the tool so that it could spin up a podcast out of it. “The team found it easy to create the podcasts without any previous experience with the tool and have been charmed by its simplicity and efficiency,” says Raphael Dardeau, chief technology officer at L’Équipe.

Using the Devoteam solution, L’Équipe was able to create all 19 podcast episodes for a series it called “2min chrono” in less than the 30 days required. Each episode is a 2-minute summary of daily Olympics highlights, covering details such as the program for the day and medal standings. The series achieved an average of 3,000 daily listeners, with a spike to 13,000 listeners.

L’Équipe now has a custom-built gen AI podcasting solution that it can continue to use for repurposing content to create podcasts at speed and to the quality standards it requires. The team also got great feedback from the internal stakeholders who were initially skeptical about the method. “The voice has really been a game changer on this project,” says Dardeau. “Our journalists and the team have been amazed at the quality of the AI speech. It has helped to spur on the internal evangelization of AI and dispelled the fears that some of our internal team members initially had about the technology.” 

L’Équipe is a French sports media outlet founded in 1946. It produces a newspaper, magazines, a website and an application, and a channel package. Its more than 350 journalists cover sports including football, cycling, motorsports, and rugby, and it has a monthly readership and viewership of 43 million readers, web users, and viewers. During the 2024 Olympics its digital platform received more than 13 million visits a day.