How Planview built a scalable AI Assistant for portfolio and project management using Amazon Bedrock

This article details how Planview built a scalable AI assistant called Planview Copilot for portfolio and project management using Amazon Bedrock, a fully managed service that provides access to foundation models. The AI assistant uses a multi-agent system architecture to handle complex queries and tasks.

Specifically, the article covers:

• The challenges faced by Planview in building the multi-agent system, such as routing tasks to appropriate agents, accessing data from various sources, and enabling self-serve creation of new AI skills
• The solution architecture, which includes an orchestrator with a router and responder component powered by large language models, and various specialized agents for application help, data queries, and executing actions
• The technical implementation using AWS services like Amazon EKS, Lambda, Amazon Bedrock, Amazon Athena, Amazon Neptune, and Amazon OpenSearch Service
• Sample prompts used for the router and responder components to demonstrate how tasks are routed and responses are generated
• The model evaluation framework and criteria used by Planview to assess performance and select the most suitable Amazon Bedrock model
• The results and impact achieved by Planview, including improved accuracy, response time, scalability, cost-efficiency, and faster time-to-market for new AI skills
• Conclusion highlighting Planview's strategies, such as modular development, robust evaluation, and Amazon Bedrock integration, that enabled the successful implementation of the AI assistant