Salesforce's AI Infrastructure team built an automated system to track AWS Bedrock model availability across 30+ AWS accounts and regions for Agentforce. Previously, engineers spent up to three days manually discovering model endpoints, checking inference profiles, and updating routing configurations. The new system uses AWS APIs queried by regional Lambda functions, publishing results as CloudWatch custom metrics visible in a unified Grafana dashboard. This cut discovery time from three days to under ten minutes, with full automation planned to make the entire discovery-to-deployment cycle nearly instantaneous. The system also implements deterministic fallback routing to verified baseline regions, preventing compliance violations from data residency requirements and avoiding outages caused by incorrect endpoint configurations.
Sort: