RETROFIT Meter Reading AI Agent

• An active AWS account
• AWS CLI installed and configured
• Permissions to create CloudFormation stacks (CAPABILITY_NAMED_IAM is required)
• Access enabled for Amazon Bedrock (AgentCore Runtime)
• Access enabled for S3 Tables
• Supported AWS Region: us-east-1

Create the stack by specifying the CloudFormation template URL.

aws cloudformation create-stack \ --stack-name meter-reading-agent-stack \ --template-url <https://cm-seller-resources.s3.us-east-1.amazonaws.com/meter-reading-ai-agent/cfn/reader-agent.yaml> \ --capabilities CAPABILITY_NAMED_IAM \ --region us-east-1 aws cloudformation describe-stacks \ --stack-name meter-reading-agent-stack \ --region us-east-1 \ --query 'Stacks[0].StackStatus'

Wait until the status becomes CREATE_COMPLETE.

Create a Markdown file describing how to read meters (e.g., needles, scales, digit interpretation), then upload it under instructions/ in the Knowledge Base source bucket.

KB_BUCKET=$(aws cloudformation describe-stacks \ --stack-name meter-reading-agent-stack \ --region us-east-1 \ --query 'Stacks[0].Outputs[?OutputKey==`KnowledgeBaseSourceBucketName`].OutputValue' \ --output text) aws s3 cp reading-instructions.md \ "s3://${KB_BUCKET}/instructions/reading-instructions.md" \ --region us-east-1

Create a JSON file that defines meter type, unit, range, graduation, and validation rules, then upload it under meter-specs/.

Example: meter-types.json

{ "meter_types": [ { "id": "pressure-gauge-001", "name": "Pressure Gauge (Analog Single Needle)", "type": "pressure", "unit": "MPa", "min_value": 0, "max_value": 10, "graduation": 0.5, "validation_rules": { "min": 0, "max": 10, "decimal_places": 2 }, "qr_code_prefix": "PRESS" } ] }

Required fields (keep descriptions concise, but do include them):

• description: Human-readable description of the meter (type, range, scale, etc.).
• reading_method: How to interpret the scale/needle and how to interpolate between marks.
• common_instructions: Shared instructions (photo angle/lighting, QR-code handling, error handling, confidence guidance).

After updating files, start a Knowledge Base ingestion job to apply the changes.

1. Upload an image to ImageBucketName (from CloudFormation Outputs).
2. The image PUT triggers EventBridge (Object Created), and Lambda automatically invokes the AgentCore Runtime Endpoint.

Processing flow:

• S3 (image PUT)
• EventBridge rule (S3 Object Created)
• Lambda (<stack>-invoke-agentcore)
• AgentCore Runtime Endpoint (qualifier: meter_reading_endpoint)

Notes:

• Supported extensions are .jpg, .jpeg, and .png (others are skipped).

On success, the agent returns the reading value, confidence, validation result, and more (example).

{ "meter_id": "001", "meter_type": "pressure", "reading_value": 5.5, "status": "processed", "reading_unit": "MPa", "confidence_score": 0.95, "validation_passed": true, "error_message": null }

For detailed deployment steps, please refer to: https://marketplace.cf.classmethod.net/services/meter-reading-ai-agent/