Construct an AI-powered automated summarization system with Amazon Bedrock and Amazon Transcribe utilizing Terraform

Extracting significant insights from unstructured information presents vital challenges for a lot of organizations. Assembly recordings, buyer interactions, and interviews include invaluable enterprise intelligence that is still largely inaccessible as a result of prohibitive time and useful resource prices of guide overview. Organizations regularly battle to effectively seize and use key data from these interactions, leading to not solely productiveness gaps but additionally missed alternatives to make use of crucial decision-making data.

This submit introduces a serverless assembly summarization system that harnesses the superior capabilities of Amazon Bedrock and Amazon Transcribe to remodel audio recordings into concise, structured, and actionable summaries. By automating this course of, organizations can reclaim numerous hours whereas ensuring key insights, motion gadgets, and choices are systematically captured and made accessible to stakeholders.

Many enterprises have standardized on infrastructure as code (IaC) practices utilizing Terraform, usually as a matter of organizational coverage. These practices are sometimes pushed by the necessity for consistency throughout environments, seamless integration with current steady integration and supply (CI/CD) pipelines, and alignment with broader DevOps methods. For these organizations, having AWS options carried out with Terraform helps them keep governance requirements whereas adopting new applied sciences. Enterprise adoption of IaC continues to develop quickly as organizations acknowledge the advantages of automated, version-controlled infrastructure deployment.

This submit addresses this want by offering an entire Terraform implementation of a serverless audio summarization system. With this resolution, organizations can deploy an AI-powered assembly summarization resolution whereas sustaining their infrastructure governance requirements. The enterprise advantages are substantial: lowered assembly follow-up time, improved data sharing, constant motion merchandise monitoring, and the flexibility to look throughout historic assembly content material. Groups can give attention to performing upon assembly outcomes relatively than struggling to doc and distribute them, driving sooner decision-making and higher organizational alignment.

What are Amazon Bedrock and Amazon Transcribe?

Amazon Bedrock is a completely managed service that provides a selection of high-performing basis fashions (FMs) from main AI corporations like AI21 Labs, Anthropic, Cohere, DeepSeek, Luma, Meta, Mistral AI, poolside (coming quickly), Stability AI, TwelveLabs (coming quickly), Author, and Amazon Nova by means of a single API, together with a broad set of capabilities to construct generative AI functions with safety, privateness, and accountable AI. With Amazon Bedrock, you possibly can experiment with and consider high FMs in your use case, customise them along with your information utilizing methods akin to fine-tuning and Retrieval Augmented Era (RAG), and construct brokers that execute duties utilizing your enterprise techniques and information sources.

Amazon Transcribe is a completely managed, automated speech recognition (ASR) service that makes it simple for builders so as to add speech to textual content capabilities to their functions. It’s powered by a next-generation, multi-billion parameter speech FM that delivers high-accuracy transcriptions for streaming and recorded speech. 1000’s of shoppers throughout industries use it to automate guide duties, unlock wealthy insights, improve accessibility, and enhance discoverability of audio and video content material.

Answer overview

Our complete audio processing system combines highly effective AWS providers to create a seamless end-to-end resolution for extracting insights from audio content material. The structure consists of two predominant parts: a user-friendly frontend interface that handles buyer interactions and file uploads, and a backend processing pipeline that transforms uncooked audio into priceless, structured data. This serverless structure facilitates scalability, reliability, and cost-effectiveness whereas delivering insightful AI-driven evaluation capabilities with out requiring specialised infrastructure administration.

The frontend workflow consists of the next steps:

1. Customers add audio recordsdata by means of a React-based frontend delivered globally utilizing Amazon CloudFront.
2. Amazon Cognito offers safe authentication and authorization for customers.
3. The applying retrieves assembly summaries and statistics by means of AWS AppSync GraphQL API, which invokes AWS Lambda capabilities to question.

The processing consists of the next steps:

1. Audio recordsdata are saved in an Amazon Easy Storage Service (Amazon S3) bucket.
2. When an audio file is uploaded to Amazon S3 within the audio/{user_id}/ prefix, an S3 occasion notification sends a message to an Amazon Easy Queue Service (Amazon SQS) queue.
3. The SQS queue triggers a Lambda operate, which initiates the processing workflow.
4. AWS Step Capabilities orchestrates all the transcription and summarization workflow with built-in error dealing with and retries.
5. Amazon Transcribe converts speech to textual content with excessive accuracy.
6. makes use of an FM (particularly Anthropic’s Claude) to generate complete, structured summaries.
7. Outcomes are saved in each Amazon S3 (uncooked information) and Amazon DynamoDB (structured information) for persistence and fast retrieval.

For extra safety, AWS Id and Entry Administration helps handle identities and entry to AWS providers and sources.

The next diagram illustrates this structure.

This structure offers a number of key advantages:

• Totally serverless – Computerized scaling and no infrastructure to handle
• Occasion-driven – Actual-time responses from parts based mostly on occasions
• Resilient – Constructed-in error dealing with and retry mechanism
• Safe – Authentication, authorization, and encryption all through
• Value-effective – Pay-per-use worth mannequin
• Globally accessible – Content material supply optimized for customers worldwide
• Extremely extensible – Seamless integration with extra providers

Let’s stroll by means of the important thing parts of our resolution in additional element.

Challenge construction

Our assembly audio summarizer undertaking follows a construction with frontend and backend parts:

sample-meeting-audio-summarizer-in-terraform/                                         
├── backend/                                                                          
│   ├── capabilities/                           # Lambda operate code                   
│   │   ├── audio-processing/                # Audio processing capabilities             
│   │   ├── authentication/                  # Authentication capabilities               
│   │   ├── data-access/                     # Knowledge entry capabilities                  
│   │   ├── queue-processing/                # SQS queue processing capabilities         
│   │   ├── summarization/                   # Summarization capabilities                
│   │   ├── transcription/                   # Transcription capabilities                
│   │   └── zipped/                          # Zipped Lambda capabilities for deployment 
│   └── terraform/                           # Infrastructure as Code                 
│       ├── modules/                         # Terraform modules                      
│       │   ├── api/                         # AppSync GraphQL API                    
│       │   ├── auth/                        # Cognito authentication                 
│       │   ├── compute/                     # Lambda capabilities                       
│       │   ├── messaging/                   # SQS queues and S3 notifications        
│       │   ├── community/                     # CloudFront and S3 web site              
│       │   ├── orchestration/               # Step Capabilities                         
│       │   ├── queue-processor/             # Queue processing Lambda                
│       │   └── storage/                     # S3 and DynamoDB                        
│       ├── predominant.tf                          # Important Terraform configuration           
│       ├── outputs.tf                       # Output values                          
│       ├── variables.tf                     # Enter variables                        
│       └── terraform.tfvars                 # Variable values                        
├── docs/                                    # Documentation and structure diagrams
├── frontend/                                # React internet utility                  
│   ├── public/                              # Public belongings                          
│   └── src/                                 # React utility supply               
│       ├── parts/                      # React parts                       
│       ├── graphql/                         # GraphQL queries and mutations          
│       ├── pages/                           # Web page parts                        
│       └── providers/                        # Service integrations                   
└── scripts/                                 # Deployment and utility scripts         
├── deploy.sh                                # Important deployment script                 
└── zip-lambdas.sh                           # Script to zip all backend lambdas  

Infrastructure setup Terraform

Our resolution makes use of Terraform to outline and provision the AWS infrastructure in a constant and repeatable method. The principle Terraform configuration orchestrates the assorted modules. The next code exhibits three of them:

# Compute Module - Lambda capabilities
module "compute" {
  supply = "./modules/compute"
  
  aws_region                        = var.aws_region
  aws_account                       = information.aws_caller_identity.present.account_id
  meeting_statistics_table_name     = var.meeting_statistics_table_name
  meeting_summaries_table_name      = var.meeting_summaries_table_name
  cognito_user_pool_id              = module.auth.cognito_user_pool_id
  iam_roles                         = module.auth.iam_roles
  storage_bucket                    = module.storage.storage_bucket
  model_id                          = var.model_id
  inference_profile_prefix          = var.inference_profile_prefix
}

# Orchestration Module - Step Capabilities
module "orchestration" {
  supply = "./modules/orchestration"
  
  aws_region                              = var.aws_region
  aws_account                             = information.aws_caller_identity.present.account_id
  storage_bucket                          = module.storage.storage_bucket
  iam_roles                               = module.auth.iam_roles
  lambda_functions                        = module.compute.lambda_functions
}

# Queue Processor Module - ProcessTranscriptionQueueFunction Lambda
module "queue_processor" {
  supply = "./modules/queue-processor"
  
  storage_bucket                    = module.storage.storage_bucket
  state_machine_arn                 = module.orchestration.state_machine_arn
  lambda_function_transcription_role = module.auth.iam_roles.lambda_function_transcription_role
  
  depends_on = [
    module.storage,
    module.orchestration
  ]
}

Audio processing workflow

The core of our resolution is a Step Capabilities workflow that orchestrates the processing of audio recordsdata. The workflow handles language detection, transcription, summarization, and notification in a resilient method with correct error dealing with.

Amazon Bedrock for summarization

The summarization part is powered by Amazon Bedrock, which offers entry to state-of-the-art FMs. Our resolution makes use of Anthropic’s Claude 3.7 Sonnet model 1 to generate complete assembly summaries:

immediate = f"""Even when it's a uncooked transcript of a gathering dialogue, missing clear construction and context and containing a number of audio system, incomplete sentences, and tangential subjects, PLEASE PROVIDE a transparent and thorough evaluation as detailed as potential of this dialog. DO NOT miss any data. CAPTURE as a lot data as potential. Use bullet factors as an alternative of dashes in your abstract.
IMPORTANT: For ALL part headers, use plain textual content with NO markdown formatting (no #, ##, **, or * symbols). Every part header needs to be in ALL CAPS adopted by a colon. For instance: "TITLE:" not "# TITLE" or "## TITLE".

CRITICAL INSTRUCTION: DO NOT use any markdown formatting symbols like #, ##, **, or * in your response, particularly for the TITLE part. The TITLE part MUST begin with "TITLE:" and never "# TITLE:" or any variation with markdown symbols.

FORMAT YOUR RESPONSE EXACTLY AS FOLLOWS:

TITLE: Give the assembly a brief title 2 or 3 phrases that's associated to the general context of the assembly, discover a distinctive identify such an organization identify or stakeholder and embody it within the title      

TYPE: Relying on the context of the assembly, the dialog, the subject, and dialogue, ALWAYS assign a kind of assembly to this abstract. Allowed Assembly sorts are: Consumer assembly, Crew assembly, Technical assembly, Coaching Session, Standing Replace, Brainstorming Session, Assessment Assembly, Exterior Stakeholder Assembly, Determination Making Assembly, and Drawback Fixing Assembly. That is essential, do not overlook this.

STAKEHOLDERS:
Present a listing of the individuals within the assembly, their firm, and their corresponding roles. If the identify will not be offered or not understood, please substitute the identify with the phrase 'Not acknowledged'. If a speaker doesn't introduce themselves, then do not embody them within the STAKEHOLDERS part.  

CONTEXT:
present a 10-15 abstract or context sentences with the next data: Important motive for contact, Decision offered, Remaining final result, contemplating all the knowledge above

MEETING OBJECTIVES:
present all of the aims or targets of the assembly. Be thorough and detailed.

CONVERSATION DETAILS:
Buyer's predominant considerations/requests
Options mentioned
Vital data verified
Choices made

KEY POINTS DISCUSSED (Elaborate on every level, if relevant):
Checklist all vital subjects and points
Vital particulars or numbers talked about
Any insurance policies or procedures defined
Particular requests or exceptions

ACTION ITEMS & NEXT STEPS (Elaborate on every level, if relevant):
What the client must do:
Fast actions required
Future steps to take
Vital dates or deadlines
What the corporate will do (Elaborate on every level, if relevant):
Processing or dealing with steps
Observe-up actions promised
Timeline for completion

ADDITIONAL NOTES (Elaborate on every level, if relevant):
Any notable points or considerations
Observe-up suggestions
Vital reminders

TECHNICAL REQUIREMENTS & RESOURCES (Elaborate on every level, if relevant):
Methods or instruments mentioned/wanted
Technical specs talked about
Required entry or permissions
Useful resource allocation particulars

Frontend implementation

The frontend is constructed with React and offers the next options:

• Person authentication and authorization utilizing Amazon Cognito
• Audio file add interface with progress indicators
• Abstract viewing with formatted sections (stakeholders, key factors, motion gadgets)
• Search performance throughout assembly summaries
• Assembly statistics visualization

The frontend communicates with the backend by means of the AWS AppSync GraphQL API, which offers a unified interface for information operations.

Safety issues

Safety is a high precedence in our resolution, which we handle with the next measures:

• Person authentication is dealt with by Amazon Cognito
• API entry is secured with Amazon Cognito person swimming pools
• S3 bucket entry is restricted to authenticated customers
• IAM roles comply with the precept of least privilege
• Knowledge is encrypted at relaxation and in transit
• Step Capabilities present safe orchestration with correct error dealing with

Advantages of utilizing Amazon Bedrock

Amazon Bedrock affords a number of key benefits for our assembly summarization system:

• Entry to state-of-the-art mannequins – Amazon Bedrock offers entry to main FMs like Anthropic’s Claude 3.7 Sonnet model 1, which delivers high-quality summarization capabilities with out the necessity to prepare customized fashions.
• Totally managed integration – Amazon Bedrock integrates seamlessly with different AWS providers, permitting for a completely serverless structure that scales mechanically with demand.
• Value-efficiency – On-Demand pricing means you solely pay for the precise processing time, making it cost-effective for variable workloads.
• Safety and compliance – Amazon Bedrock maintains information privateness and safety, ensuring delicate assembly content material stays protected inside your AWS setting.
• Customizable prompts – The power to craft detailed prompts permits for tailor-made summaries that extract precisely the knowledge your group wants from conferences. Amazon Bedrock additionally offers immediate administration and optimization, in addition to the playground for fast prototyping.
• Multilingual support – Amazon Bedrock can course of content material in a number of languages, making it appropriate for international organizations.
• Diminished improvement time – Pre-trained fashions reduce the necessity for in depth AI improvement experience and infrastructure.
• Steady enchancment – Amazon Bedrock offers a mannequin selection, and the person can replace the prevailing fashions with a single string change.

Stipulations

Earlier than implementing this resolution, ensure you have:

Within the following sections, we stroll by means of the steps to deploy the assembly audio summarizer resolution.

Clone the repository

First, clone the repository containing the Terraform code:

git clone https://github.com/aws-samples/sample-meeting-audio-summarizer-in-terraform
cd sample-meeting-audio-summarizer-in-terraform

Configure AWS credentials

Make sure that your AWS credentials are correctly configured. You should utilize the AWS CLI to arrange your credentials:

aws configure --profile meeting-summarizer

You can be prompted to enter your AWS entry key ID, secret entry key, default AWS Area, and output format.

Set up frontend dependencies

To arrange the frontend improvement setting, navigate to the frontend listing and set up the required dependencies:

cd frontend
npm set up

Create configuration recordsdata

Transfer to the terraform listing:

cd ../backend/terraform/  

Replace the terraform.tfvars file within the backend/terraform listing along with your particular values. This configuration provides values for the variables beforehand outlined within the variables.tf file.

You’ll be able to customise different variables outlined in variables.tf in line with your wants. Within the terraform.tfvars file, you present precise values for the variables declared in variables.tf, so you possibly can customise the deployment with out modifying the core configuration recordsdata:

aws_region                              = "us-east-1"                                  
aws_profile                             = "YOUR-AWS-PROFILE"                           
setting                             = "prod"                                       
app_name                                = "meeting-audio-summarizer"                   
dynamodb_read_capacity                  = 5                                            
dynamodb_write_capacity                 = 5                                            
cognito_allowed_email_domains           = ["example.com"]                              
model_id                                = "anthropic.claude-3-7-sonnet-20250219-v1:0"  
inference_profile_prefix                = "us"                                         
frontend_bucket_name                    = "a-unique-bucket-name"                       
storage_bucket                          = "a-unique-bucket-name"                       
cognito_domain_prefix                   = "meeting-summarizer"                         
meeting_statistics_table_name           = "MeetingStatistics"                          
meeting_summaries_table_name            = "MeetingSummaries"  

For a-unique-bucket-name, select a novel identify that’s significant and is sensible to you.

Initialize and apply Terraform

Navigate to the terraform listing and initialize the Terraform setting:

terraform init

To improve the beforehand chosen plugins to the latest model that complies with the configuration’s model constraints, use the next command:

terraform init -upgrade

It will trigger Terraform to disregard picks recorded within the dependency lock file and take the latest accessible model matching the configured model constraints.

Assessment the deliberate modifications:

terraform plan

Apply the Terraform configuration to create the sources:

terraform apply

When prompted, enter sure to substantiate the deployment. You’ll be able to run terraform apply -auto-approve to skip the approval query.

Deploy the answer

After the backend deployment is full, deploy all the resolution utilizing the offered deployment script:

cd ../../scripts
sudo chmod +x deploy.sh
./deploy.sh

This script handles all the deployment course of, together with:

• Deploying the backend infrastructure utilizing Terraform
• Robotically configuring the frontend with backend useful resource data
• Constructing and deploying the frontend utility
• Establishing CloudFront distribution
• Invalidating the CloudFront cache to ensure the newest content material is served

Confirm the deployment

After all the resolution (each backend and frontend) is deployed, in your terminal it is best to see one thing just like the next textual content:

Deployment full! :)

============================================================================
Your app is out there at: https://d1e5vh2t5qryy2.cloudfront.web.
============================================================================

The CloudFront URL (*.cloudfront.web/) is exclusive, so yours won’t be the identical.

Enter the URL into your browser to open the applying. You will notice a login web page like the next screenshot. You need to create an account to entry the applying.

Begin by importing a file:

View generated summaries in a structured format:

See assembly statistics:

Clear up

To cleanup the answer you will need to run this command.

terraform destroy

This command will utterly take away the AWS sources provisioned by Terraform in your setting. When executed, it would show an in depth plan exhibiting the sources that shall be destroyed, and immediate for affirmation earlier than continuing. The method might take a number of minutes because it systematically removes infrastructure parts within the right dependency order.

Bear in mind to confirm the destruction is full by checking your AWS Console to ensure no billable sources stay energetic.

Value issues

When implementing this resolution, it’s vital to know the price implications of every part. Let’s analyze the prices based mostly on a sensible utilization state of affairs, based mostly on the next assumptions:

• 50 hours of audio processing monthly
• Common assembly size of half-hour
• 100 energetic customers accessing the system
• 5 million API queries monthly

Nearly all of the price comes from Amazon Transcribe (roughly 73% of whole price at $72.00), with AWS AppSync being the second largest price part (roughly 20% at $20.00). Regardless of offering the core AI performance, Amazon Bedrock prices roughly 3% of whole at $3.00, and DynamoDB, CloudFront, Lambda, Step Capabilities, Amazon SQS, and Amazon S3 make up the remaining 4%.

We are able to make the most of the next price optimization alternatives:

• Implement audio compression to cut back storage and processing prices
• Use Amazon Transcribe Medical for medical conferences (if relevant) for increased accuracy
• Implement caching methods for regularly accessed summaries to cut back AppSync and DynamoDB prices
• Take into account reserved capability for DynamoDB if utilization patterns are predictable

The next desk summarizes these costs. Refer the AWS pricing pages for every service to study extra in regards to the AWS pricing mannequin.

Subsequent steps

The following part of our assembly summarization resolution will incorporate a number of superior AI applied sciences to ship larger enterprise worth. Amazon Sonic Mannequin can enhance transcription accuracy by higher dealing with a number of audio system, accents, and technical terminology—addressing a key ache level for international organizations with various groups. In the meantime, Amazon Bedrock Flows can improve the system’s analytical capabilities by implementing automated assembly categorization, role-based abstract customization, and integration with company data bases to supply related context. These enhancements will help organizations extract actionable insights that will in any other case stay buried in dialog.

The addition of real-time processing capabilities helps groups see key factors, motion gadgets, and choices as they emerge throughout conferences, enabling fast clarification and lowering follow-up questions. Enhanced analytics performance observe patterns throughout a number of conferences over time, giving administration visibility into communication effectiveness, decision-making processes, and undertaking progress. By integrating with current productiveness instruments like calendars, every day agenda, activity administration techniques, and communication providers, this resolution makes certain that assembly intelligence flows immediately into every day workflows, minimizing guide switch of data and ensuring crucial insights drive tangible enterprise outcomes throughout departments.

Conclusion

Our assembly audio summarizer combines AWS serverless applied sciences with generative AI to resolve a crucial productiveness problem. It mechanically transcribes and summarizes conferences, saving organizations 1000’s of hours whereas ensuring insights and motion gadgets are systematically captured and shared with stakeholders.

The serverless structure scales effortlessly with fluctuating assembly volumes, prices simply $0.98 per assembly on common, and minimizes infrastructure administration and upkeep overhead. Amazon Bedrock offers enterprise-grade AI capabilities with out requiring specialised machine studying experience or vital improvement sources, and the Terraform-based infrastructure as code permits fast deployment throughout environments, customization to satisfy particular organizational necessities, and seamless integration with current CI/CD pipelines.

As the sector of generative AI continues to evolve and new, better-performing fashions develop into accessible, the answer’s means to carry out its duties will mechanically enhance on efficiency and accuracy with out extra improvement effort, enhancing summarization high quality, language understanding, and contextual consciousness. This makes the assembly audio summarizer an more and more priceless asset for contemporary companies trying to optimize assembly workflows, improve data sharing, and enhance organizational productiveness.

Further sources

Discuss with Amazon Bedrock Documentation for extra particulars on mannequin choice, immediate engineering, and API integration in your generative AI functions. Moreover, see Amazon Transcribe Documentation for details about the speech-to-text service’s options, language help, and customization choices for attaining correct audio transcription. For infrastructure deployment wants, see Terraform AWS Supplier Documentation for detailed explanations of useful resource sorts, attributes, and configuration choices for provisioning AWS sources programmatically. To boost your infrastructure administration abilities, see Finest practices for utilizing the Terraform AWS Supplier, the place you will discover really useful approaches for module group, state administration, safety configurations, and useful resource naming conventions that may assist ensure your AWS infrastructure deployments stay scalable and maintainable.

Concerning the authors

Dunieski Otano is a Options Architect at Amazon Internet Companies based mostly out of Miami, Florida. He works with World Broad Public Sector MNO (Multi-Worldwide Organizations) prospects. His ardour is Safety, Machine Studying and Synthetic Intelligence, and Serverless. He works along with his prospects to assist them construct and deploy excessive accessible, scalable, and safe options. Dunieski holds 14 AWS certifications and is an AWS Golden Jacket recipient. In his free time, you’ll find him spending time along with his household and canine, watching an important film, coding, or flying his drone.

Joel Asante, an Austin-based Options Architect at Amazon Internet Companies (AWS), works with GovTech (Authorities Expertise) prospects. With a powerful background in information science and utility improvement, he brings deep technical experience to creating safe and scalable cloud architectures for his prospects. Joel is obsessed with information analytics, machine studying, and robotics, leveraging his improvement expertise to design revolutionary options that meet advanced authorities necessities. He holds 13 AWS certifications and enjoys household time, health, and cheering for the Kansas Metropolis Chiefs and Los Angeles Lakers in his spare time.

Ezzel Mohammed is a Options Architect at Amazon Internet Companies (AWS) based mostly in Dallas, Texas. He works on the Worldwide Organizations crew inside the World Broad Public Sector, collaborating carefully with UN companies to ship revolutionary cloud options. With a Laptop Science background, Ezzeldien brings deep technical experience in system design, serving to prospects architect and deploy extremely accessible and scalable options that meet worldwide compliance necessities. He holds 9 AWS certifications and is obsessed with making use of AI Engineering and Machine Studying to handle international challenges. In his free time, he enjoys happening walks, watching soccer with family and friends, taking part in volleyball, and studying tech articles.