What was the impact of the AI solution?

Claude Sonnet 4.5 achieved about 76% internal accuracy on core Capella iQ workflows. The architecture met latency and throughput targets with no user-impacting quality regressions during controlled traffic testing. Model upgrades or provider changes can be made through configuration rather than code changes.

Source

ExploringEvidence: Medium50/100

Couchbase builds Capella iQ developer assistant on Amazon Bedrock

Use case typeDeveloper productivityUpdated Jul 20, 2026

Couchbase built Capella iQ as an AI-powered developer assistant that generates SQL++ queries, recommends indexes, and supports multi-turn conversations. The implementation uses a model-agnostic inference architecture on Amazon Bedrock with Amazon Elastic Kubernetes Service, private Amazon VPC connectivity, and Cross-Region Inference across AWS Regions for resilience and burst handling.

Organization: Couchbase
Industry: Tech & Comms
Location: United States
Published: July 2026

Reported outcomes

+76%

internal accuracy on core Capella iQ workflowsQuality & accuracy

Strategic outcomes

Speed & agilityModel changes can be made through configuration instead of code changesScale & capacityHandled bursty workloads with automatic cross-region routing

Why do we believe this?Outcome claims, sources, and evidence checks

Normalized claim

Internal accuracy on core Capella iQ workflows: 76% increase

AWS Machine Learning BlogJul 20, 2026Blog postExplicit claimMedium evidence strength

Claude Sonnet 4.5 achieved approximately 76 percent accuracy on an internal evaluation modeled on

Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks

Customer: Couchbase
Provider: AWS
Maturity: Exploring
Linked source: AWS Machine Learning Blog

The team built automated benchmarking and evaluation pipelines to compare models on functional correctness, determinism, latency, and formatting consistency

Customer identity supportedSource describes one deploymentMaturity supported

Primary read

Use case focus

Showing 3 of 3

1Developer productivity
2Conversational assistants
3Workflow orchestration

Enterprise adoption of Capella iQ required flexible foundation model support, high availability across AWS Regions, and bursty scaling without pre-provisioned capacity.

Couchbase integrated Amazon Bedrock into a multi-model inference architecture hosted on Amazon Elastic Kubernetes Service.
Requests pass through Amazon VPC interface endpoints to the Bedrock runtime, with Cross-Region Inference and tenant-level routing and configuration to avoid code changes.
The team built automated benchmarking and evaluation pipelines to compare models on functional correctness, determinism, latency, and formatting consistency.

Technologies

Amazon Bedrock Amazon Elastic Kubernetes Service Amazon VPC Cross-Region Inference Anthropic Claude Sonnet 4.0

Claude Sonnet 4.5 achieved about 76% internal accuracy on core Capella iQ workflows.
The architecture met latency and throughput targets with no user-impacting quality regressions during controlled traffic testing.
Model upgrades or provider changes can be made through configuration rather than code changes.

Architecture

Capella iQ runs on an Amazon Elastic Kubernetes Service control plane across two AWS Regions. The cp-api, cp-internal-api, and cp-ns microservices orchestrate model routing and tenant-level configuration. Inference requests flow privately through an Amazon VPC interface endpoint to Amazon Bedrock runtime, which uses Cross-Region Inference across us-east-1, us-east-2, and us-west-2 for automatic failover and load distribution. The team also built automated evaluation pipelines and custom test harnesses for benchmarking and failover validation.

Sources & evidence1

Evidence: Medium50/100Evidence strength