Restaurant.buzz

Inspiration

Choosing a restaurant usually means looking at food, price, location, and ratings. But one thing is surprisingly hard to know before you arrive: will I actually be able to hear the person across the table? Yes, this is what my Mom says to me every time...

Restaurant.buzz started with a simple idea: restaurant data and reviews already contain thousands of clues about noise, atmosphere, seating, crowds, music, and conversation. What if AI could turn those clues into something useful?

What it does

Restaurant.buzz helps diners discover restaurants based on predicted noise level and conversation-friendliness, not just star ratings.

The project combines restaurant data with review analysis to create a noise score. Instead of simply asking an AI model whether a restaurant is "loud," the system looks for multiple signals: references to music, crowds, acoustics, packed rooms, conversation difficulty, quietness, seating, and other contextual clues. And it uses those signals to produce a more consistent prediction.

The goal is to make questions like "Where can four people have dinner and actually talk?" searchable.

How we built it

We started with NYC Open Data, using the city's restaurant dataset as the foundation for building a searchable universe of New York City restaurants.

From there, we enriched restaurant records with additional information and review data from sources including Google Places and Yelp.

Restaurant.buzz was built in CHATGPT and Google AI Studio using Gemini, with a noise-prediction engine that combines:

• Structured restaurant information
• Review text and sentiment
• Noise-related keywords and contextual signals
• AI classification of ambiguous review language
• A weighted scoring system that converts those signals into an understandable noise prediction

The interface was designed to make the resulting score useful rather than technical, allowing diners to quickly understand what kind of environment they should expect.

Challenges

The biggest challenge was that noise is subjective. One person's "energetic" restaurant is another person's "impossibly loud" restaurant. That meant we couldn't rely on keyword counting alone. We needed both a scoring framework and AI interpretation to understand the context behind those comments. We also had to work quickly across multiple APIs, data formats, rate limits, and incomplete restaurant records during the hackathon.

What we learned

One of the most interesting discoveries was how much more useful restaurant data becomes when different sources are combined.

NYC Open Data provides a broad, structured foundation of restaurants. Google Places adds richer business information and customer reviews. Yelp contributes another set of restaurant and review signals. Gemini can then analyze the unstructured language across those sources and turn it into additional attributes that aren't available in any single dataset. Ideally, It's about connecting different kinds of data to create information that didn't previously exist.

What's next

The next step is to make the noise model even more accurate by combining predicted noise with real-world sound measurements. Restaurant.buzz could incorporate noise readings collected by audiologists and other acoustic experts, while also allowing diners to submit their own measurements and experiences. Over time, those measurements could help validate and improve the AI model while revealing how noise changes by time of day, day of the week, seating area, and crowd level.

Built With

• ai/ml
• chatgpt
• gemini-api
• google-ai-studio
• google-cloud
• google-places
• javascript
• nyc-open-data
• tailwind-css
• yelp-fusion-api