DEMO / RECORDING
Overview & Result
Santiago demonstrates filtering hotel reviews with Jev. Apify collects the reviews, TypeSafe Jev classifies Wi-Fi and noise mentions, and GPT-5 mini reviews uncertain cases. Python calculates the ranking; GPT explains the result. The public repository documents the pipeline, but we have not independently run it.
How Jev fits in the loop
- Apify collects text reviews for the hotels listed in HOTELS.
- The script calls typesafe/jev-1.13 through OpenRouter to classify Wi-Fi and noise mentions. Reviews with an unclear label, confidence below 0.9, or an API error go to GPT-5 mini.
- Python ranks hotels using smoothed positive/complaint counts, with Wi-Fi and noise weighted equally. GPT explains the computed result without changing the ranking.
How to reproduce
- Open Santiago’s original video and the linked hotel-reviews repository.
- Follow the README: Python 3.9+, APIFY_TOKEN and OPENROUTER_API_KEY are required for collection and classification. Edit HOTELS for your destination.
- Use a small review sample first. Compare the labels with your own reading, including uncertain and unresolved reviews.
Repository setup and a small sample (README)
git clone https://github.com/svpino/hotel-reviews.git
cd hotel-reviews
python3 hotels.py --reviews 100Why this build matters
Turn a large review pile into a shortlist based on explicit preferences, while keeping uncertain evidence visible.
Limitations
- Repository and integration inspected; no independent end-to-end execution or accuracy benchmark.
- No mention of Wi-Fi or noise is not positive evidence. Sparse samples and unresolved classifications affect the result.
- The score is a ranking rule, not the probability that a hotel is good. Price and location are outside this demo.
- Apify and OpenRouter usage are billed separately. The saved-review mode still makes model calls.