AWS Builder Center
InfraShift: Predicting AWS Infrastructure Impact Before Deployment

InfraShift: Predicting AWS Infrastructure Impact Before Deployment

Building an AI-powered pre-deployment intelligence platform for safer, smarter infrastructure changes

Modern applications evolve quickly. A single code change can affect multiple AWS resources, increase infrastructure costs, change performance characteristics, or introduce unexpected operational risks.
During the AWS Hackathon Builder challenge, our team explored this problem and built InfraShift — a platform designed to provide infrastructure impact intelligence before a change is deployed.

The Problem

When a developer creates a Pull Request, the focus is usually on whether the application code works.
But infrastructure changes can have a much larger impact.
A seemingly small change can potentially:
  • Increase AWS resource usage
  • Change application performance
  • Increase infrastructure cost
  • Affect dependent resources
  • Introduce operational risks
  • Behave differently from previous deployments
The challenge is that developers often don't have all the historical and infrastructure context required to understand these effects before deployment.
We wanted to answer a simple question:
"What could happen to my infrastructure if I deploy this change?"
That question became the foundation of InfraShift.



What is InfraShift?

InfraShift is a pre-deployment infrastructure impact intelligence platform.
Instead of waiting for a deployment to discover its consequences, InfraShift analyzes a proposed change and provides information about:
  • AWS resources that may be affected
  • Historical evidence
  • Infrastructure telemetry
  • Expected cost changes
  • Expected performance changes
  • Confidence and assumptions
  • Recommendations before deployment
  • Actual results after deployment
The overall workflow is:
GitHub Pull Request
↓
Feature Extraction
↓
AWS Resource Impact
↓
Historical Evidence + Telemetry
↓
Cost / Performance Forecast
↓
Recommendations
↓
Deployment
↓
Actual Outcome
↓
Feedback Loop
This creates a continuous learning cycle rather than treating deployment as the end of the process.



How InfraShift Works

1. Start with a Pull Request

The process begins with a GitHub Pull Request.
Instead of requiring developers to manually inspect infrastructure, InfraShift uses the proposed change as the starting point for analysis.
The frontend allows users to provide a GitHub repository and Pull Request number.
For example:
Repository:
https://github.com/owner/repository

Pull Request:
123
InfraShift can then construct the corresponding Pull Request URL:
https://github.com/owner/repository/pull/123
The design is repository-independent so that the workflow isn't restricted to a single project.



2. Feature Extraction

The next stage is understanding what changed.
The system extracts relevant characteristics from the Pull Request and converts the raw code change into information that can be analyzed.
The goal isn't simply to display a Git diff.
Instead, the change becomes a set of meaningful signals that can be connected to infrastructure impact.



3. AWS Resource Impact

Once the change is understood, InfraShift determines which infrastructure resources may be affected.
For example, a code change could potentially affect:
Application
↓
Compute
↓
Database
↓
Network
↓
Storage
The Impact Graph provides a visual representation of these relationships.
This makes it easier for developers to understand the potential blast radius of a change instead of examining infrastructure components independently.



4. Historical Evidence and Telemetry

Predictions become more useful when they are supported by evidence.
InfraShift therefore separates information into different categories:

Observation

Something that has actually happened in historical data.

Prediction

Something the system expects to happen after deployment.

Inference

A conclusion derived from available observations and relationships.
This distinction is important because a prediction should not be presented as if it were an established fact.
InfraShift can surface historical comparisons, telemetry information, assumptions, and confidence alongside the forecast.



5. Cost and Performance Forecast

The system then generates a forecast for potential infrastructure effects.
The forecast can focus on areas such as:
  • Infrastructure cost
  • Resource utilization
  • Performance
  • Potential operational impact
Instead of presenting a prediction without context, the interface provides supporting evidence and confidence information.
This allows developers to understand not only what the forecast is, but also why the system produced it.



6. Recommendations

After analyzing the potential impact, InfraShift provides recommendations that developers can review before deployment.
The recommendation interface is designed to answer:
"What should I consider before deploying this change?"
Recommendations can be evaluated using supporting evidence, assumptions, and validation information.
This turns the platform from a monitoring dashboard into a decision-support workflow.



7. Deployment and Outcome Validation

One of the most important parts of InfraShift is that the system does not stop after making a prediction.
After deployment, actual results can be compared with the original forecast.
For example:
Forecast
↓
Deployment
↓
Actual Result
↓
Compare
↓
Forecast Error
The Outcome page allows users to examine:
  • Forecast vs actual values
  • Cost differences
  • Performance differences
  • Forecast error
  • Validation timeline
  • Supporting evidence
This creates the foundation for improving future forecasts.



The Feedback Loop

The complete InfraShift lifecycle can therefore be represented as:
┌──────────────────────┐
│ GitHub Pull Request│
└──────────┬───────────┘
↓
┌──────────────────────┐
│ Feature Extraction │
└──────────┬───────────┘
↓
┌──────────────────────┐
│ AWS Impact Analysis │
└──────────┬───────────┘
↓
┌──────────────────────┐
│ Evidence + Telemetry │
└──────────┬───────────┘
↓
┌──────────────────────┐
│ Cost/Performance │
│ Forecast │
└──────────┬───────────┘
↓
┌──────────────────────┐
│ Recommendations │
└──────────┬───────────┘
↓
Deployment
↓
┌──────────────────────┐
│ Actual Outcome │
└──────────┬───────────┘
↓
┌──────────────────────┐
│ Feedback Loop │
└──────────────────────┘
The important idea is that historical outcomes can become evidence for future analysis.



Frontend Experience

We built the frontend around a workflow rather than a collection of disconnected dashboards.
The main areas include:

Dashboard

Provides an overview of infrastructure analyses and their current state.

PR Analysis

Allows users to start an analysis from a Pull Request.

Analysis Details

Provides a detailed view of the analyzed change.

Impact Graph

Visualizes potentially affected infrastructure resources.

Forecast

Displays projected cost and performance impact.

Evidence

Explains the evidence behind the forecast.

Recommendations

Provides pre-deployment recommendations.

Outcome

Compares predictions with actual deployment results.

History

Allows users to review previous analyses.
This structure helps users move from:
Change → Impact → Evidence → Forecast → Recommendation → Outcome



Why Explainability Matters

One of the design principles behind InfraShift is that predictions should not appear as unexplained numbers.
For example, instead of simply showing:
Expected Cost: +18%
the system should help answer:
Why?

Historical evidence
+
Telemetry
+
Infrastructure relationships
+
Assumptions
=
Forecast
This is particularly important for infrastructure decisions because users need context before acting on a forecast.



Challenges We Faced

Building InfraShift required combining several different areas:
  • Frontend development
  • Backend APIs
  • AWS infrastructure
  • AI-powered analysis
  • GitHub integration
  • Historical data
  • Forecasting
  • Visualization
One challenge was designing an interface that could present complex infrastructure information without overwhelming the user.
Another challenge was maintaining a clear distinction between:
what the system knows, what it predicts, and what it infers.
We also designed the project using separate team areas so that frontend, backend, AWS infrastructure, and AI components could be developed independently.



What We Learned

Through this project, we learned that infrastructure intelligence isn't only about predicting a number.
A useful system needs to connect several pieces:
Code Change
+
Infrastructure Context
+
Historical Evidence
+
Telemetry
+
Forecast
+
Actual Outcome
We also learned the importance of making predictions explainable and validating them against real-world results.



Future Scope

There are several directions in which InfraShift could be extended.

Automated Infrastructure Detection

Automatically identify infrastructure resources affected by a Pull Request.

More AWS Services

Expand analysis across additional AWS services and resource relationships.

Better Forecasting

Use larger historical datasets and improved forecasting techniques.

Continuous Learning

Use actual deployment outcomes to improve future predictions.

CI/CD Integration

Integrate InfraShift directly into CI/CD pipelines so infrastructure impact analysis becomes part of the development workflow.

Automated Policy Checks

Add configurable organizational policies for cost, performance, and infrastructure risk.



Conclusion

InfraShift explores a different approach to infrastructure management.
Instead of asking only:
"Did the deployment succeed?"
we want to help teams ask earlier:
"What impact could this change have before we deploy it?"
By connecting GitHub changes, AWS infrastructure, historical evidence, telemetry, forecasting, recommendations, and post-deployment outcomes, InfraShift creates a continuous pre-deployment intelligence and feedback loop.
The goal is not to replace developers or DevOps teams.
It is to give them more context before infrastructure changes reach production.



Built for the AWS Hackathon Builder Challenge

Project: InfraShift Focus: Pre-deployment infrastructure impact intelligence Repository:  swapnil1222589/InfraShift 
Team: InfraShift
 AWS   Generative AI   DevOps   Cloud Computing   Infrastructure   AI   GitHub   Serverless   AWS Builder 
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article