preflight

Your AI co-pilot that detects and fixes system problems before your software ships

Comment

Inspiration

Teams at startups and hackathons often move very quickly, which means projects sometimes get launched with problems that were easy to miss: exposed API keys, localhost URLs still in the code, unfinished UI text, weak admin routes, or visual issues that make the product feel unfinished.

These problems usually appear right before demo day or launch, when there is very little time left to review everything manually.

We wanted to build a system that acts like a final pre-launch checkpoint — something that reviews both the code and the user interface and answers one simple question:

Is this project actually ready to launch?

This idea became Preflight.

What it does

Preflight is a multimodal launch-readiness agent built for the Amazon Nova Hackathon.

It analyzes a project by reviewing:

  • a ZIP archive of the project
  • a public GitHub repository
  • an optional screenshot of the UI

After scanning the project, Preflight generates a launch readiness report that includes:

  • a launch readiness score
  • a final verdict on whether the project is launch-ready
  • critical blockers and warnings
  • UI issues detected from screenshots
  • step-by-step suggestions to fix problems
  • an optional voice summary for quick review

We also built a CLI version so developers can run the same analysis directly from their terminal as part of their local workflow.

How we built it

The frontend was built using Next.js and React, with a dashboard interface where users can:

  • upload a project ZIP
  • paste a GitHub repository URL
  • run a demo scan
  • view the launch report
  • listen to the voice summary

The backend uses Node.js and Express to handle:

  • file uploads
  • ZIP extraction
  • repository cloning
  • rule-based static scanning
  • report generation
  • API communication with Amazon Nova

For AI reasoning we used Amazon Bedrock with Amazon Nova, which enables multimodal analysis of both code and UI screenshots.

The scanning pipeline works in several stages:

  1. Static analysis to detect common risky patterns
  2. Multimodal AI reasoning over project files and screenshots
  3. launch-readiness scoring
  4. generation of clear remediation steps
  5. voice output for quick executive-style summaries

We also included a built-in demo mode to make the system easy to test during judging.

Challenges we ran into

One of the biggest challenges was combining different types of signals into one clear decision.

A project might look fine in the code but still have obvious UI problems, or the opposite might be true.

We had to design the system so that:

  • code findings
  • screenshot analysis
  • risk verification
  • final scoring

all contributed to one coherent launch report.

Another challenge was designing a launch score that actually means something. We wanted Preflight to be conservative so that a project is only marked as launch-ready when it truly appears safe to deploy.

We also spent time improving the report interface so the results would be clear, readable, and easy for judges to understand during a demo.

What we learned

During this project we learned that AI tools are much more useful when they provide clear decisions instead of just analysis.

Rather than listing many possible issues, Preflight focuses on delivering:

  • a clear verdict
  • a readiness score
  • actionable steps to fix problems

We also learned that hackathon projects become easier to demonstrate when they support multiple workflows such as:

  • a web interface
  • GitHub repository scanning
  • a demo mode
  • a command-line interface

These entry points made the project feel more practical and complete.

Accomplishments that we're proud of

We are proud that Preflight goes beyond a simple code scanner.

The system combines:

  • code review
  • visual UI analysis
  • remediation planning
  • voice summaries
  • web and CLI workflows

into a single tool.

We are also proud of the stricter launch-gate logic. The “Ready to Launch” verdict is intentionally difficult to achieve, which makes the results more realistic and trustworthy.

What's next for Preflight

In the future we would like to add:

  • GitHub PR and CI/CD integration
  • deeper security and authentication checks
  • multi-screen screenshot analysis
  • release-history comparisons
  • team dashboards
  • automatic fix suggestions or pull requests

Our long-term vision is for Preflight to become the final check teams run before launching a product, presenting at demo day, or releasing software to real users.

Built With

Share this project:

Updates