Explainable first
If we can’t show the sentence and phrases behind a flag, we don’t ship the flag.
About
Built by a student team, held to a product standard.
HomeLens AI started as the final project for CIS 509, Analytics for Unstructured Data. We wanted to see how far free-form customer language could go as an early-warning signal for service quality.
Mission
Make complaints legible.
Home-service customers describe problems in their own words. We turn that language into signals a manager can read, challenge and act on, without pretending the signal is a verdict.
Careful with claims
Reviews are allegations. Our language stays at “risk signal” and “management attention,” never “fraud” or “liability.”
Privacy by default
Providers are anonymized, text is masked before modeling and pasted reviews are not stored.
Team
The people behind it.
Rithik Roy ThatiCIS 509 project team
Kanha JodhpurkarCIS 509 project team
Sankalp Sharma MadgulaCIS 509 project team
Dev BhattacharyyaCIS 509 project team
Under the hood
Technology.
Modeling
Fine-tuned DistilBERT, TF–IDF logistic regression, a lexicon baseline, and a phrase-dictionary aspect extractor.
Data
Yelp Open Dataset reviews of home-service providers in Tucson, AZ, processed in a reproducible notebook pipeline.
Product
A Streamlit command center for analysts and this Next.js site, deployed on Vercel, which runs the same analysis in TypeScript.
