Building a Multi-Agent System with Amazon Bedrock for Business Insights

In the rapidly evolving world of business, the ability to gather and analyze data from multiple sources is crucial. This is especially true for the biopharmaceutical industry, where companies need to understand drug market sizes, clinical trial outcomes, and legal and financial implications to make informed decisions. Amazon Bedrock Agents and its multi-agent collaboration feature offer a powerful solution to these challenges.

Business intelligence and market research are essential for capturing industry trends and shaping key business strategies. Biopharmaceutical companies, for instance, use data to analyze drug market sizes, clinical trials, side effects, and patent and legal briefs to form investment strategies. However, accessing and analyzing information scattered across multiple data sources can be complex and time-consuming. Consolidating and querying these disparate datasets requires navigating different data formats and access mechanisms, which can be a significant barrier.

Generative AI agentic systems have emerged as a promising solution, enabling organizations to use AI for autonomous reasoning and action-based tasks. However, many of these systems are built with a single-agent setup, which can be limiting in a complex business environment. Managing multiple data sources and encoding information for various business domains can lead to prompt bloat, causing the large language model (LLM) to suffer from 'forgetting the middle' of long contexts. Additionally, using a single LLM can limit cost, latency, and accuracy optimizations.

Amazon Bedrock Agents and its multi-agent collaboration feature provide an enterprise-ready solution for these challenges. As a managed service within the AWS ecosystem, Amazon Bedrock Agents offers seamless integration with AWS data sources, built-in security controls, and enterprise-grade scalability. It also supports additional features like Amazon Bedrock Guardrails and Knowledge Bases, reducing deployment overhead and empowering developers to focus on agent logic.

To demonstrate the capabilities of multi-agent collaboration, consider a fictional biopharmaceutical company, PharmaCorp. PharmaCorp faces the challenge of managing vast amounts of data across its R&D, Legal, and Finance divisions. Each division works with structured data, such as stock prices, and unstructured data, such as clinical trial reports. The data for each division is stored in different data stores, making it difficult to access cross-functional insights and slowing down decision-making processes.

To address this, PharmaCorp builds a multi-agent system with domain-specific sub-agents for each division. The main agent acts as an orchestrator, asking questions to multiple sub-agents and synthesizing retrieved data. This setup empowers PharmaCorp to access expertise and information within minutes that would otherwise take hours of human effort to compile.

The R&D sub-agent has access to clinical trial data through Amazon Athena and unstructured clinical trial reports. The legal sub-agent has access to unstructured patents and lawsuit legal briefs. The finance sub-agent has access to research budget data through Athena and historical stock price data stored in Amazon Redshift. Each sub-agent has granular permissions to access only the data in its domain.

The synthetic data used in this example is generated using Anthropic’s Claude 3.5 Sonnet model and encompasses three domains: R&D, Legal, and Finance. The data is aligned within and across domains to ensure consistency. For example, clinical trial reports map to trial data, and stock prices correlate with patents and lawsuits.

The Pharma R&D domain includes tables for Drugs, Drug Trials, and Side Effects, queried from Amazon S3 through Athena. The unstructured data consists of synthetic clinical trial reports with detailed information about trial design, outcomes, and recommendations.

The Legal domain has unstructured data comprising patents and lawsuit legal briefs, containing information about invention backgrounds, descriptions, experimental results, and court proceedings.

The Finance domain includes tables for Stock Price and Research Budgets. The Stock Price table is stored in Amazon Redshift and contains historical monthly stock prices and volume. The Research Budgets table is accessed from Amazon S3 through Athena and contains annual budgets for each department.

The multi-agent framework showcases how data from multiple sources and databases can be synthesized. In practice, data could also be stored in other databases such as Amazon RDS.

Anthropic’s Claude 3 Sonnet is used in this multi-agent demonstration due to its balance of intelligence and speed. Depending on the use case, a more intelligent or smaller, faster model can be employed.

To deploy this solution, you need an active AWS account and access to Amazon Titan Embeddings G1 – Text, Anthropic’s Claude 3 Sonnet, and Anthropic’s Claude 3.5 Sonnet on Amazon Bedrock. The deployment process involves using AWS CloudFormation to create resources such as S3 buckets, AWS Lambda functions, an Amazon Bedrock agent, and a knowledge base.

By leveraging Amazon Bedrock Agents and multi-agent collaboration, organizations can break down data silos, enhance cross-functional insights, and make informed decisions more efficiently.