FairFlow AI

Inspiration

Online purchases often begin with a low advertised price but end with hidden fees, preselected add-ons, or unexpected subscriptions. Existing detectors usually inspect only one page, even though these problems often emerge across several checkout steps. We built FairFlow AI to show users the true cost and conditions before they pay.

What it does

FairFlow is a privacy-conscious browser assistant that compares each stage of an online purchase. It detects:

• Newly added or hidden fees
• Preselected services and insurance
• Free trials that become paid subscriptions
• Visually weakened reject or cancel options
• Differences between the advertised and final price

It displays an evidence-based timeline showing what changed, where it changed, and how much extra the user may pay.

[ \text{Price Increase}= \frac{P_{\text{final}}-P_{\text{initial}}} {P_{\text{initial}}}\times100\% ]

How we built it

The browser extension extracts visible text, prices, checkbox states, and interface layout while excluding sensitive payment fields. A hybrid ML system combines:

• Text and price extraction
• Visual hierarchy analysis
• Consent and checkbox detection
• Sequential comparison between pages
• Evidence-based risk scoring

Unlike a single-screenshot classifier, FairFlow treats checkout as a sequence of page states and identifies changes between them.

Challenges

The main challenge was distinguishing deceptive design from legitimate fees or clearly explained subscriptions. We addressed this by analyzing context, visual prominence, and when information appeared during checkout.

Another challenge was generalizing across websites with different layouts. To avoid template memorization, we evaluate FairFlow on websites not included in its training data.

What we learned

We learned that deceptive design cannot be reliably detected through keywords alone. The same words may be harmless when clearly presented and harmful when hidden or combined with default consent.

We also learned that consumer-protection AI must be explainable. FairFlow therefore shows the evidence behind every warning and labels uncertain cases instead of making unsupported accusations.

What’s next

We plan to expand FairFlow to travel booking, mobile purchases, multilingual interfaces, and subscription cancellation flows.

FairFlow reveals hidden fees, unwanted subscriptions, and deceptive design before users click “Pay.”

Built With

• chrome
• docker
• fastapi
• hugging-face
• onnx
• opencv
• paddleocr
• playwright
• python
• pytorch
• react
• scikit-learn
• tailwind-css
• typescript
• vite