BrumBuster

BrumBuster protects car part buyers from innaccurate and faked listings by cross referencing photos across the web and using AI Vision for verification. Now the steering wheels back in your hands.

Comment

Inspiration

Buying car parts online is currently a trust minefield. Scammers frequently list parts using manufacturer stock illustrations, or use listing images from completely different vehicles to create "ghost listings." We wanted to build a tool that shifts the power back into the buyer's hands, allowing everyday drivers to spot this before spending their hard earned cash.

What it does

BrumBuster is an automated visual forensics platform for the online automotive marketplace. A user simply pastes an eBay UK listing URL and selects an image index from the gallery to run a deep forensic audit.

The platform immediately delivers a unified dashboard tracking:

  • Match Verification: An AI evaluation of physical details (like lug patterns and part layout) against the listing title to catch mismatches or deceptive item conditions.
  • Duplicates Tracker: An exact match asset scan across the global web index to map out unauthorized asset duplicates, foreign hosting domains, and significant time discrepencies.

How we built it

We engineered BrumBuster using a split full-stack architecture:

  • The Backend: A Python Flask server that handles multi-image web scraping and HTML parsing.
  • The Forensic Engines: We integrated SerpApi’s Google Lens API exact matches clustering to isolate duplicates. We then queried Gemini 3 Flash to provide Automotive Forensics.
  • The Frontend: A responsive React application styled with Tailwind CSS

Challenges we ran into

  • CDN Web Scraping Failures: Passing raw image URLs directly to cloud vision APIs triggered immediate bot protection blocks from image hosts. We solved this by updating our backend pipeline to scrape, download, and pipe the image directly through local memory as a byte stream.
  • Visual Generalization Noise: Standard visual searches flag all shiny brake discs or silver alloy rims as identical, generating massive false positive data, so we constrained our pipeline to SerpApi's exact asset matrix.
  • Handling Stock Images: Detecting them whilst avoiding them accidentally masking a physical component mismatch or defect using logical overrides.

Accomplishments that we're proud of

  • Precision Asset Auditing: Getting Gemini 3 Flash to successfully identify highly specific mechanical details, like a bolt pattern on a brake disc and cross referencing it with vehicle specifications.
  • Dynamic Gallery Indexing: Giving users the control to bypass a glossy stock cover photo and inspect hidden secondary images where scams are usually obscured.

What we learned

We gained deep experience in web scraping, but also learned the massive difference between image similarity and exact file duplicates in computer vision workflows.

What's next for BrumBuster

We also want to scale our custom scraping workers to support a wider array of regional automotive platforms, building a comprehensive web extension safety layer for digital buyers.

Built With

Share this project:

Updates