RentReady Vision for Fire TV

RentReady Vision turns rental walkthrough videos into AI-assisted inspection findings, with room-by-room evidence review and Amazon Bedrock-powered readiness briefings on Fire TV.

Comment 1

Inspiration

Rental property inspections often involve lengthy walkthrough videos, numerous photographs, and findings that can be difficult to review and communicate. Property owners and managers need an easier way to identify potential maintenance issues, examine supporting evidence, and discuss what needs attention.

I created RentReady Vision to transform ordinary walkthrough videos into structured, AI-assisted inspection reports. For the Amazon Developer Hackathon, I wanted to bring those findings to the largest screen in the home: Fire TV.

What it does

RentReady Vision analyzes rental property walkthrough videos using computer vision and AI to identify potential issues, organize inspection evidence, and generate structured findings.

The new Fire TV experience allows users to:

  • Navigate inspections and rooms using a TV remote.
  • Review potential maintenance findings room by room.
  • Examine preserved photographic evidence on a large screen.
  • Play selected video clips associated with findings.
  • Distinguish human-confirmed findings from rejected AI suggestions.
  • View an Amazon Bedrock-powered rental-readiness briefing with prioritized recommendations.

The goal is to make inspection results easier to understand, verify, and discuss collaboratively.

Findings are advisory and require human verification; the application is not a substitute for a professional property inspection.

How I built it

The existing RentReady Vision backend uses Python, FastAPI, OpenCV, and AWS infrastructure to process uploaded walkthrough videos.

OpenCV evaluates video frames for sharpness, exposure, motion, scene changes, and duplication. Selected evidence frames support downstream AI analysis and inspection reporting.

The backend uses Amazon S3 for private media storage, Amazon DynamoDB for inspection state, Amazon SQS for asynchronous processing, and Amazon EC2 with Graviton-based processing infrastructure.

For the hackathon, I developed a new Android/Kotlin application targeting the Fire TV experience.

The television interface uses remote-first navigation, Android TV components, and authenticated API integrations to retrieve saved inspections, evidence photographs, and video clips.

I also integrated Amazon Bedrock using Amazon Nova 2 Lite to generate rental-readiness recommendations from saved inspection findings. Recommendations are validated and cached in DynamoDB before being presented on the television.

The Fire TV application accesses cached inspection results rather than processing walkthrough videos directly on the television.

Challenges I encountered

Several challenges required particular attention:

  • Designing predictable D-pad navigation and focus behavior for a television remote.
  • Maintaining secure access to private inspection images and video evidence.
  • Implementing restricted device pairing without exposing privileged AWS credentials.
  • Preserving the relationship between AI-generated recommendations, original findings, and subsequent human-review decisions.
  • Controlling AWS processing costs through on-demand infrastructure and cached model results.
  • Establishing a reproducible Android TV emulator testing and judge-access workflow.

Accomplishments

I developed a working television application that extends RentReady Vision beyond its original browser-based inspection workflow.

The application supports room navigation, evidence review, video playback, restricted access to an approved inspection, and a Bedrock-powered readiness briefing.

I validated five consecutive end-to-end demonstration journeys on a Google Android TV emulator, including a cold boot, and prepared both an offline demonstration and a restricted live AWS demonstration for judges.

The Android TV emulator is an accepted hackathon testing environment. Physical Fire TV hardware and Fire OS compatibility have not been independently tested.

What I learned

This project reinforced the importance of designing AI applications around verifiable evidence and human judgment rather than treating model predictions as unquestionable conclusions.

I also gained practical experience with television-first interaction design, Android/Kotlin development, secured device authentication, AWS service integration, and managing cloud costs.

What's next

Potential future improvements include collecting reviewer feedback on false positives and missed issues, expanding detector evaluation, improving inspection accuracy, and adding broader device compatibility testing.

Significant updates during the hackathon

RentReady Vision development began on August 26, 2026, five days before the Amazon Developer Hackathon started.

The initial project was an early OpenCV video-processing prototype. Development of the inspection platform continued during the competition.

Beginning October 11, I developed the Amazon-specific television application, including its remote-first interface, room-by-room inspection review, evidence playback, secure pairing, and Amazon Bedrock-powered briefing.

These additions created a new television-based inspection experience while reusing the existing computer-vision and AWS backend.

The development history and distinction between existing functionality and new hackathon work are documented in the project's GitHub repository.

Built With

Share this project:

Updates

Submission history