Questions, answered plainly.
What is HomeLens AI?
A risk command center that turns customer reviews of home-service providers into review-level risk signals, explainable complaint aspects, provider monitoring and recommended actions for a person to review.
Does it detect fraud or predict that a business will fail?
No. It reports review-based risk signals and relative concern tiers. It does not verify any allegation, establish misconduct or liability, or predict bankruptcy. See Responsible AI.
How accurate is it?
On held-out businesses, the fine-tuned DistilBERT reaches 0.973 macro-F1 and the TF–IDF logistic regression 0.951. Scores are for classifying low-star versus high-star reviews, not for judging a business. Details on How it works.
Which model powers the website demo?
The TF–IDF logistic regression, ported to TypeScript so it runs inside this site. DistilBERT is the primary model in the notebook and Streamlit app; it is not bundled here because of its size.
What data does it use?
Yelp Open Dataset reviews of home-service providers in Tucson, AZ, from November 2006 to January 2022. Providers are anonymized as Provider_XXXX and need at least 20 reviews to be monitored.
Can I use it on my own reviews?
You can paste any review into the analyzer. Uploading a full export is on the roadmap. Note the models were trained on Yelp reviews from one market.
Is my pasted text stored?
No. It is processed for the request and returned. See the privacy policy.
Why is a review flagged but no aspect is shown?
The model found the review high-risk but no dictionary phrase matched. HomeLens routes these to manual review instead of guessing an aspect.
What do the tiers mean?
Stable below 50, Watch 50–75, Elevated 75–90, High concern 90 and above. They rank providers against each other in this corpus and are not probabilities.
Is it free?
The demo is free with no account. There are no paid plans today. See Pricing.
