Building Guardian Life’s First Production-Scale AI Assistant

Challenge

Rearc worked directly with Guardian Life, a large insurance provider, to design and deploy a production-scale AI assistant for customer support agents, internal product teams, and third-party sales representatives. The assistant needed to answer complex policy questions across multiple product domains by searching policy details, cross-referencing internal product documentation, and synthesizing accurate responses in natural language.

This was the organization’s first major AI chatbot deployed at scale, requiring new patterns for secure LLM access to enterprise APIs, product data, and regulated insurance information.

Solution

• AI: Leveraging Claude Sonnet on AWS Bedrock, Rearc built an agentic assistant capable of routing user questions, synthesizing policy information, and coordinating multiple tools through natural language.
• Search: Rearc integrated AWS Knowledge Bases, reranking, and specialized search tools to retrieve policy details, product rules, documentation, and regulatory context across life and disability insurance domains.
• MCP: Rearc deployed the organization’s first production MCP server, establishing reusable patterns for how enterprise APIs should expose tools, schemas, and permissions to LLM-powered applications.
• Security: Rearc designed authorization patterns that separated platform-level permissions from user-level privileges, enforcing access controls directly inside the tool layer before sensitive systems were exposed to the agent.
• Data: Rearc transformed complex enterprise API schemas into LLM-friendly tool interfaces, removing sensitive data and reshaping bloated payloads into structures the model could reliably reason over.

Outcome

• Rearc delivered the organization’s first production-scale AI chatbot, serving internal product teams, customer support agents, and third-party sales representatives.
• The assistant reduced the cognitive burden of answering complex insurance questions by synthesizing information across large policy datasets, dense product documentation, business rules, and difficult-to-interpret API outputs.
• The assistant replaced fragmented search workflows with a natural-language interface that could infer query parameters from conversation, route requests to the right policy or product data sources, and return context-aware answers without requiring users to navigate multiple systems.
• The project established the organization’s first production MCP patterns, creating a secure model for connecting LLMs to enterprise APIs through permissioned, schema-controlled tools.
• Rearc created reusable authorization and tool-design patterns for agentic applications, including user-context enforcement, sensitive-data filtering, and LLM-optimized API schemas.