The Smart Prescription Reader is a prototype solution that extracts structured data from prescription images using AWS Bedrock foundation models. It employs a multi-stage AI approach with specialized models for extraction, evaluation, and correction, ensuring high accuracy while optimizing for cost and performance. Built on serverless AWS services including Step Functions, Lambda, and AppSync, the solution provides a scalable architecture with built-in security controls for handling sensitive medical information. This prototype demonstrates how generative AI can be practically applied to automate document processing while maintaining accuracy through automated quality checks and corrections.
The Smart Prescription Reader employs a multi-stage AI-driven process to extract structured data from prescription images. It uses generative AI foundation models for initial extraction, quality evaluation, and intelligent correction, balancing speed, accuracy, and cost-effectiveness. The system adapts to varying prescription complexities, optimizes performance through techniques like prompt caching, and maintains high accuracy with built-in quality assurance mechanisms. This approach ensures scalable, auditable, and flexible prescription data extraction that can evolve with business needs.
View Detailed Technical Approach.
You are responsible for the cost of the AWS services used while running this solution. As of April 2025, the cost for running this solution with the default settings in the US East (N. Virginia) Region is approximately $3,173.69 per month processing 75,000 prescriptions.
The costs shown represent pricing without consideration of AWS free tier benefits.
We recommend creating a budget through AWS Cost Explorer to help manage costs. Prices are subject to change. For full details, refer to the pricing webpage for each AWS service used in this solution.
The following table provides a sample cost breakdown for deploying this solution with the default parameters in the US East (N. Virginia) Region for one month:
| AWS Service | Usage | Cost [USD] |
|---|---|---|
| Amazon Textract | Extract unstructured text from images | $112.50 |
| Amazon Bedrock | Extract prescriptions(Anthropic Claude 3 Haiku) | $206.25 |
| Amazon Bedrock | Evaluate result(Amazon Nova Pro) | $918.00 |
| Amazon Bedrock | Correct result(Anthropic Claude 3.7 Sonnet) | $1,845.00 |
| AWS Step Functions | Workflow orchestration | $7.88 |
| AWS Lambda | Function compute | $25.15 |
| Amazon DynamoDB | Job status storage | $1.85 |
| Amazon S3 | Image and config storage | $0.79 |
| AWS AppSync | GraphQL API | $8.00 |
| AWS CloudWatch | Logs | $7.57 |
| AWS KMS | Encryption keys | $10.00 |
| AWS PrivateLink | VPC endpoints | $29.30 |
| Other services | WAF, Cognito | $1.40 |
| Total | $3,173.69 |
With prompt caching enabled for supported models, the Bedrock costs can be reduced considerably.
Cost per prescription:
- With correction: $0.0423
- Without correction: $0.0116
The actual cost will vary based on:
- Number of prescriptions processed
- Percentage requiring corrections (default estimate assumes 80%)
- Length and complexity of prescriptions
- Selected Bedrock models
- Whether prompt caching is enabled
This project is a monorepo managed by Nx with packages for infrastructure, business logic, and the demo application.
| Folder | Package |
|---|---|
| packages/infra/ | Infrastructure as Code in AWS CDK |
| packages/core/ | Business logic in Python used in AWS Lambda functions |
| packages/demo/ | React demo application with Cloudscape components |
| packages/common/models-graphql/ | GraphQL schema and generated types |
- Configure the AWS Credentials in your environment. Refer to Authentication and access.
- Download and install AWS CLI. Refer to Installing the AWS CLI.
- Install and configure AWS CDK. Refer to Installing the AWS CDK.
- Download and install Docker. Refer to Docker.
- NodeJS >= 18.0.0. You can manage NodeJs versions with nvm or pnpm
- Python >= 3.13. You can manage Python versions with uv.
- uv >= 0.6.0. Refer to uv installation guide.
- pnpm >= 8.6.3. Refer to pnpm installation guide.
# Install pnpm
curl -fsSL https://get.pnpm.io/install.sh | sh -
# Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh
# Install required tools and versions
pnpm env use --global 22
uv python install 3.13
# Install global packages
npm install -g aws-cdk
# Bootstrap AWS CDK
cdk bootstrapThis prototype can use the Amazon Nova and Anthropic Claude models on Amazon Bedrock. You will need to enable them for your account in the regions you want to use.
If you choose to use Cross-Region Inference for increased throughput, you will need to activate the models in each region that will be used.
The entire build and deployment can be done with just two commands:
pnpm install
nx deploy @smart-prescription-reader/infraThe deploy command will automatically trigger the necessary builds and then deploy the infrastructure. You should see output similar to:
SmartPrescriptionReader: creating CloudFormation changeset...
✅ SmartPrescriptionReader
✨ Deployment time: 68.35s
Outputs:
SmartPrescriptionReader.ConfigBucketName = smartprescriptionreader-graph-configbucket2112c5ec-nklpf9njwtb3
SmartPrescriptionReader.ImagesBucketName = smartprescriptionreader-graph-imagesbucket1e86afb2-gj37da0zaxgv
SmartPrescriptionReader.UserIdentityPoolId = us-east-1_bQtq1lmbk
SmartPrescriptionReader.UserIdentityUserIdentityIdentityPoolId7D1EA400 = us-east-1:548af07e-e760-4391-b4e4-46783af2d6e7
SmartPrescriptionReader.UserIdentityUserIdentityUserPoolId21B35D31 = us-east-1_bQtq1lmbk
SmartPrescriptionReader.WebsiteUrl = See packages/demo/README.md
SmartPrescriptionReader.WorkflowCorrectResponseLambdaCorrectResponseLambdaConfigC73411FF = /SmartPrescriptionReader/CorrectResponseLambdaConfig
SmartPrescriptionReader.WorkflowEvaluateResponseLambdaEvaluateResponseLambdaConfigA5909445 = /SmartPrescriptionReader/EvaluateResponseLambdaConfig
SmartPrescriptionReader.WorkflowExtractPrescriptionLambdaExtractPrescriptionLambdaConfig28E0C5F9 = /SmartPrescriptionReader/ExtractPrescriptionLambdaConfig
After deploying the infrastructure, you can optionally configure the following components to enhance the Smart Prescription Reader's performance:
- Upload Medication List
Upload a list of medications to the configuration bucket to improve medication name recognition. Depending on your use case, this can include trade names, active ingredients, and dosage.
aws s3 cp path/to/your/medication-list.txt s3://<ConfigBucketName>/medications/drug-list.txt- Upload Glossary
Upload a glossary of medical terms to the configuration bucket to enhance term recognition.
aws s3 cp path/to/your/glossary.txt s3://<ConfigBucketName>/glossary/prescription-terms.txt- Configure Lambda Functions
Use AWS Systems Manager Parameter Store to configure the extraction, evaluation, and correction Lambda functions. The parameter names are provided in the CloudFormation stack outputs.
The configuration supports prompt caching for supported Bedrock models to improve performance and reduce costs. As of April 2025, prompt caching is supported for:
- Anthropic Claude 3.7 Sonnet (anthropic.claude-3-7-sonnet-20250219-v1:0)
- Anthropic Claude 3.5 Haiku (anthropic.claude-3-5-haiku-20241022-v1:0)
- Amazon Nova Micro (amazon.nova-micro-v1:0)
- Amazon Nova Lite (amazon.nova-lite-v1:0)
- Amazon Nova Pro (amazon.nova-pro-v1:0)
Example configuration (adjust values as needed):
aws ssm put-parameter \
--name "/SmartPrescriptionReader/ExtractPrescriptionLambdaConfig" \
--type "String" \
--value '{
"modelId": "anthropic.claude-3-haiku-20240307-v1:0",
"temperature": 0.0,
"medicationsKey": "medications/drug-list.txt",
"glossaryKey": "glossary/prescription-terms.txt",
"thinking": true,
"transcribe": false,
"promptCacheModels": [
"anthropic.claude-3-7-sonnet-20250219-v1:0",
"anthropic.claude-3-5-haiku-20241022-v1:0",
"amazon.nova-micro-v1:0",
"amazon.nova-lite-v1:0",
"amazon.nova-pro-v1:0"
]
}' \
--overwrite
aws ssm put-parameter \
--name "/SmartPrescriptionReader/EvaluateResponseLambdaConfig" \
--type "String" \
--value '{
"modelId": "anthropic.claude-3-sonnet-20240229-v1:0",
"temperature": 0.0,
"medicationsKey": "medications/drug-list.txt",
"glossaryKey": "glossary/prescription-terms.txt",
"thinking": true,
"transcribe": false,
"promptCacheModels": [
"anthropic.claude-3-7-sonnet-20250219-v1:0",
"anthropic.claude-3-5-haiku-20241022-v1:0",
"amazon.nova-micro-v1:0",
"amazon.nova-lite-v1:0",
"amazon.nova-pro-v1:0"
]
}' \
--overwrite
aws ssm put-parameter \
--name "/SmartPrescriptionReader/CorrectResponseLambdaConfig" \
--type "String" \
--value '{
"modelId": "anthropic.claude-3-sonnet-20240229-v1:0",
"temperature": 0.0,
"medicationsKey": "medications/drug-list.txt",
"glossaryKey": "glossary/prescription-terms.txt",
"thinking": true,
"transcribe": false,
"promptCacheModels": [
"anthropic.claude-3-7-sonnet-20250219-v1:0",
"anthropic.claude-3-5-haiku-20241022-v1:0",
"amazon.nova-micro-v1:0",
"amazon.nova-lite-v1:0",
"amazon.nova-pro-v1:0"
]
}' \
--overwriteAdjust the configuration values based on your specific requirements and the available Bedrock models in your region. The
promptCacheModels list can be updated as more models add support for prompt caching.
See Amazon Bedrock prompt caching documentation
for the latest list of supported models.
These configuration steps are optional but can significantly improve the accuracy and performance of the Smart Prescription Reader.
The Smart Prescription Reader includes a demo web application in packages/demo that showcases the prescription data extraction capabilities. The demo allows you to:
- Upload prescription images
- Review extracted data alongside the original image
In the event that you decide to stop using the prototype, we recommend that you follow a tear-down procedure. Most of the services used have no cost when there is no active use with the notable exception of storage in S3, DynamoDB, and CloudWatch Logs and keys in AWS KMS. AWS CloudFormation via CDK will tear down all resources except for Amazon S3 buckets and AWS CloudWatch Logs log groups with data and AWS KMS Customer Managed Keys(CMK).
- On the AWS CloudFormation console or using AWS CDK in the terminal, destroy the stacks that were deployed. Some of the S3 buckets will remain as they will not be empty.
- Delete any CloudWatch Logs log groups that you do not wish to keep.
- In any S3 buckets that remain that you do not wish to keep, empty the buckets by disabling logging and configuring a lifecycle policy that expires objects after one day. Wait a day.
- After a day, go back and delete the buckets.
- Schedule deletion of any KMS CMKs that you do not wish to keep.
As part of the prototype process, we have some improvement ideas for your path to production.
Please see the security guidelines.
This is sample code, for non-production usage. Before using Smart Prescription Reader, you should consider conducting your own independent assessment to ensure that your use would comply with your own specific security and quality control practices and standards, as well as the laws, rules, and regulations that govern you and your content.