Cara on AWS: domain-specific AI for enterprise brokerages

Cara has built a domain-specific AI platform on AWS that automates back-office processes for enterprise insurance brokerages, addressing manual workflows and a talent shortage in an $8 trillion industry. The system runs on Amazon EKS for orchestration and uses Amazon Bedrock-hosted foundation models for inference, with tenant-isolated workspaces and customer integrations.

How is Cara built on AWS?

Cara runs its microservices on Amazon Elastic Kubernetes Service across multiple Availability Zones and uses Amazon Bedrock for LLM inference. EKS manages ingestion pipelines, workflow engines, and the inference layer, while Bedrock provides access to foundation models via a managed API so Cara does not manage GPU infrastructure directly.

The deployment model provisions parameterized templates per tenant, creating isolated namespaces, storage, and inference endpoints. Security features cited include tenant data isolation, encryption at rest and in transit, and integration with AWS Identity and Access Management for enterprise controls. Cara also integrates with agency management systems and CRM tools to sync accounts, policies, and documents.

What insurance workflows does Cara automate?

Cara automates coverage and quote intelligence, application and form automation, proposal and renewal generation, and knowledge-driven workflows that reference agency guidelines and historical placements. The platform cross-fills ACORD and supplemental forms from source documents and prior submissions, compares carrier quotes and highlights exclusions, and produces branded proposals and renewal spreadsheets.

Workflows operate directly inside existing broker technology stacks to reduce duplicate data entry. The product’s design goal is fast time-to-value: enterprise brokerages can be onboarded within hours, and customized workflows can go live within days using the tenant provisioning templates on EKS.

What measurable outcomes has Cara delivered?

Cara reports concrete operational results: about 10 hours saved per user per week through workflow automation and contextual knowledge retrieval, onboarding completed within hours, custom workflows live within days, support for thousands of concurrent users and workflows per brokerage, and adoption by hundreds of leading insurance agencies and brokerages.

Those metrics are driven by organisation-specific workflow automation and domain-specific AI that understands carrier requirements and brokerage processes. The platform’s elastic scaling and Kubernetes Horizontal Pod Autoscaler adjust capacity to support peak renewal and servicing periods.

Why it matters

Insurance processes involve sensitive PII, financial records, and regulatory constraints that generic AI tools do not model out of the box. Cara’s approach pairs domain-specific models and workflow logic with tenant-isolated, auditable deployments on AWS, which addresses precision, compliance, and security requirements that brokerages demand. The combination targets a fundamental bottleneck: agents spending hours on repetitive tasks such as re-keying data and completing applications while the industry faces a persistent talent shortage.

What to watch

Watch whether Cara expands the number of AI-driven workflows beyond sales and renewals into broader servicing and operations, and whether reported adoption grows past the current claim of hundreds of agencies. Also track additional integrations with major AMS and CRM vendors and any changes to how inference is hosted or scaled on Bedrock and EKS.

"We are thrilled to advance the boundaries of domain-specific AI in real-world insurance use cases with AWS," said Vic Yeh, CEO of Cara.

Technical summary

Nodes: Amazon EKS, Amazon Bedrock, ingestion pipelines, workflow engines, inference layer, tenant workspaces, storage, AMS/CRM integrations, AWS IAM.

Cara deploys multi-AZ EKS clusters, isolates each organization in its own namespace, uses Bedrock-hosted foundation models for inference, and syncs data with existing agency management and CRM systems to minimise agent workflow changes.