AI Workshop - Boomi + AWS Overview and Comments

 

The AWS + Boomi Innovation Lab was a hands-on workshop that showcased how enterprises can use Boomi and Amazon Bedrock to orchestrate AI workflows that tackle real-world data quality challenges. The session walked participants through building an intelligent support case assistant using Boomi as the integration layer, Salesforce and ServiceNow as systems of record, and Amazon Bedrock to power a generative AI agent.

In this post,. I'll cover the workshop content and tools, the role of the different components, orchestration vs MCP, testing and validation concerns, and additional resources.

The goal was practical: demonstrate how to build an AI agent that ingests support case data, reasons over it using foundation models (FMs), and returns validated, actionable responses. Participants followed a structured exercise to build this pipeline—from data ingestion in Salesforce to automated insights and ticket creation in ServiceNow. The workshop emphasized real-world relevance, demonstrating how to ground generative outputs in enterprise data using Retrieval Augmented Generation (RAG).

Boomi served as the orchestration engine—detecting new Salesforce cases, packaging them as structured XML payloads, and routing those to Amazon Bedrock. Once the AI agent returned control with actionable tasks, Boomi executed updates across Salesforce and ServiceNow via pre-built connectors and low-code flows.

Salesforce was the origin of the data—the workshop simulated a new support case with various product and issue details.

Amazon Bedrock acted as the cognitive layer. It accessed Knowledge Bases (S3-hosted PDFs, FAQs, and structured Salesforce records), interpreted the case context using Claude 3.5 Sonnet, and issued recommendations and actions. See the Bedrock summary PDF below for a sense of the configuration options - they seem to be very broad and customizable.

ServiceNow was the endpoint for task creation, illustrating how AI insights can drive automated workflows in Case Management tools.

Amazon Bedrock supports model choice (Claude, Llama, Amazon Titan, etc.) via a single API. In the lab, Bedrock used a Knowledge Base with product data to support RAG-enhanced responses. Embeddings were generated and indexed in a vector store, allowing semantic search to inform prompts dynamically. Importantly, no customer data is used for model training—everything remains within your VPC, supporting enterprise-grade privacy and compliance (HIPAA, SOC, GDPR).

Boomi played the central role of workflow manager. It invoked the Bedrock agent, received structured outputs, and executed those outputs via Salesforce and ServiceNow function groups. Boomi’s visual interface made it easy to sequence multi-step flows and enforce logic between steps. Bedrock agents were treated as intelligent sub-processes—reasoning on their own but deferring to Boomi for execution.

The newest concept in AI is MCP - Model Context Protocol. It is just a more composable, pluggable, discoverable way to connect resources and services to an AI model. In this workshop, Boomi took on the role of injecting context and case specifics into the prompt, and parsing the output for actionable data, and orchestrating the followup actions.

AI-generated outputs were subjected to guardrails and contextual grounding. Topics like PII and Salesforce financials were masked or denied. Outputs were validated for relevance and traceability. This is crucial when using AI in enterprise settings—validation steps must be explicit, inspectable, and testable. That said, testing and verification is a major gap. None of my customers are ready to trust an untested AI Agent with anything more important than marketing copy and personalized sales scripts.

I started my career as a Biochemist working on clinical trial dat for a major Pharma company. Clinical trials confront a similar problem to AI - testing safety and efficacy in a very complex black box system - the human body. In a future post, I'll draw the parallels, and suggest how lessons from Clinical Trials can inform testing and validation in AI systems.

You can explore the full workshop materials, code samples, and build instructions in the AWS workshop - sign up here. Or you can see a ten minute video of the functioning agent in this Youtube LINK . For an in-depth look at Amazon Bedrock’s capabilities, check the [Bedrock Overview PDF] or visit AWS documentation. Boomi partners and customers can find additional resources and templates in the Boomi community portal.