Thomson Reuters Democratizes AI to Fuel Innovation Using Amazon Bedrock

Thomson Reuters is a global provider of trusted content and technology to professionals across the legal, tax, and media industries. Thomson Reuters Labs, the company’s dedicated innovation team, has been pioneering AI and natural language processing solutions for over 30 years. The company now uses AI to improve information retrieval and automate content-sourcing processes.

Using AWS, Thomson Reuters developed Open Arena and, subsequently, its generative AI platform, a web-based playground where users can experiment with a growing set of ML tools powered by large language models (LLMs). The concept was born out of the company’s desire to explore and innovate across all segments of the organization, making AI solutions accessible to both technical and nontechnical teams. Using the platform’s chat-based interface, employees without coding backgrounds can explore and develop solutions while learning techniques and best practices. The benefits include more scalable delivery; considerations for security, privacy, and ethics; and change management.

“We built Open Arena to give our colleagues a safe and transparent way of understanding this new, massive technology shift and how they could apply it to solving customers’ challenges,” says Joel Hron, head of AI and Thomson Reuters Labs at Thomson Reuters. “Using AWS, we can now efficiently use our world-class content and expertise with more breadth and efficiency than we could have before.”

Initially, its generative AI platform was used for existing generative AI applications, but Thomson Reuters saw an opportunity to expand users’ selection of LLMs beyond what was already available by incorporating Amazon Bedrock into Open Arena.

“We chose Amazon Bedrock for accessibility because we can seamlessly integrate our content with a variety of models,” says Hron. “Having the ability to use a diverse range of models as they come out was a key driver for us, especially given how quickly this space is evolving.”

With the extensive selection of powerful foundation models on Amazon Bedrock, Thomson Reuters can customize the choice of models teams can try for different content and products. The solution helps abstract some of the complexity for the user, making it easier to choose the best option for a particular use case. Thomson Reuters has also accelerated AI model deployment times using Amazon Bedrock. Now, development teams can access models using Amazon Bedrock in minutes or hours instead of days or weeks. This additional flexibility and agility make it seamless to integrate content into other AWS services used by Thomson Reuters.

“The ease of spinning up new use cases is really powerful,” says Hron. “In a quickly evolving space, you want to diversify and be able to capitalize on the latest advancements as they come. Using Amazon Bedrock, we can find the right model for the right job.”

Thomson Reuters’s product teams used Amazon Bedrock to build a tax research generative AI application called Checkpoint Edge with CoCounsel. It can provide responsive answers to customer queries with in-line citations to tax resources. This dialogue-centric application joins the Thomson Reuters’s line of Checkpoint tax solutions as its first generative AI search product.

“When initially developing Checkpoint Edge with CoCounsel, we didn’t need to burden our engineering teams, which was a huge benefit,” says Hron. “Instead, the product teams could independently explore in Open Arena using LLMs on Amazon Bedrock to see if this was a viable idea.”

Checkpoint Edge with CoCounsel is just one example of how Thomson Reuters’s generative AI platform has simplified the user experience in developing generative AI solutions. The ability to experiment on different use cases opens a wide range of opportunities. Thomson Reuters employees can simply drop data into Open Arena, try out a few prompts, and see what other types of similar applications teams across the organization have developed.

“Generative AI has democratized accessibility to AI in a way we haven’t seen before,” says Hron. “Many project concepts can be tested in Open Arena without requiring a significant amount of training data or a significant level of ML expertise, which opens up a channel for every employee to explore new ideas without constraints.”