Sentinel - Secure Voice-Governed Incident Response for AWS

Inspiration

💡 Inspiration

Operational incident response teams frequently suffer from two opposing problems: "alert fatigue" leading to ignored warnings, and "autonomous agent risk," where giving an AI language model unrestricted access to run raw infrastructure code results in catastrophic downtime or massive billing anomalies. We built Sentinel to resolve this exact paradigm—providing a secure, voice-governed, human-in-the-loop control plane that safely converts chaotic alerts into deterministic, voice-approved infrastructure fixes.

⚙️ What it does

Sentinel acts as a highly structured security layer for AWS environments. When an operational metric triggers a critical incident:

1. It ingests and redacts telemetry text to protect system security.
2. It uses Amazon Bedrock to securely map the error onto a closed, pre-approved runbook catalog.
3. It presents the unified diagnostic evaluation to an engineer via an Alexa+ Model Context Protocol (MCP) conversation.
4. It enforces cryptographic parameter verification and transactional DynamoDB concurrency blocks to completely eliminate execution locks or double-triggers before sending an execution call to AWS Systems Manager (SSM) Automation.

🛠️ How we built it

• Infrastructure Architecture: Structured cleanly via an AWS SAM / CloudFormation template deploying single-table DynamoDB patterns.
• Triage Layer: Programmed in Python 3.11 using native Boto3 Bedrock clients to constrain model parameters and enforce strict allowlisted behaviors.
• Concurrency Layer: Implemented custom Python modules mapping SHA-256 parameter action digests paired with DynamoDB Conditional Update Expressions to secure execution flow boundaries.
• Automation Execution: Integrated native, robust AWS SSM UpdateService API pipelines targeted at predictable ECS Fargate Task rolling redeployments.

🏆 Track & Mini-Challenge Nominations

• Primary Track - Alexa+: We have built a fully simulated self-hosted MCP server runtime architecture to smoothly expose functional incident tooling scopes securely to the Alexa+ workspace without needing raw hardware attachments.
• Mini-Challenge - AWS Builder: This application features extensive, multi-layered usage of native AWS core resources, featuring Amazon Bedrock text-classification modeling, Amazon DynamoDB transactional logic tracking, and AWS Systems Manager infrastructure document actions.

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for Sentinel - Secure Voice-Governed Incident Response for AWS

Built With

• amazon-bedrock
• amazon-dynamodb
• aws-lambda
• aws-systems-manager
• cloudformation
• github
• python