This article describes how Thomson Reuters developed Open Arena, an enterprise-grade LLM playground, in under 6 weeks using AWS services.

• Open Arena is a web-based playground enabling non-technical employees to experiment with LLMs safely
• Built using AWS Lambda, DynamoDB, SageMaker, and Hugging Face Deep Learning Containers
• Features modular serverless architecture with REST API integration for flexible model deployment
• Includes pre-set tiles for specific use cases: open-source LLM chat, document Q&A, text summarization
• Implements retrieval-augmented generation (RAG) pipeline for content-based queries using OpenSearch
• Achieved over 1,000 monthly internal users within one month of launch
• Sparked enterprise-wide innovation with employees discovering new AI use cases for their workflows
• Secured data remains encrypted within VPC during fine-tuning and inference processes

Open Arena successfully democratized LLM access across Thomson Reuters, fostering company-wide AI experimentation and innovation while maintaining security and ease of use.