SF Eats

What to eat & where in San Francisco. Pick a cuisine, pick a dish, see the best spots for that dish — ranked, mapped, and scored.

Every spot is scored on two signals:

• Hype — how many Google reviews mention the dish by name
• Love — what percentage of those reviews are positive (4-5 stars or positive sentiment)

When a spot has high hype but low love, it gets flagged as a hype trap so you don't waste a meal.

How it works

A static HTML front-end (Leaflet map + panels) loads pre-generated data from sf-food-data.js. The data is built by a Node CLI that queries Google Places API for each dish, pulls full review lists via SerpAPI, and computes hype/love scores using keyword matching + sentiment analysis. No framework, no build step.

Community recommendations are stored in a Convex backend — anyone viewing the app can submit their pick for the best spot for any dish, and it shows up for everyone immediately.

Cuisines

Currently covers Mexican, Italian, and Turkish dishes:

Mexican	Italian	Turkish
Mission Burrito	Margherita Pizza	Manti
Carne Asada	Focaccia	Lahmacun
Quesabirria	Cacio e Pepe	Pide
Al Pastor	Bolognese	Turkish Breakfast
Carnitas	Squid Ink Pasta	Borek
Ceviche & Aguachile	Gnocchi	Adana Kebab
Tamales	Lasagna	Doner Kebab
Mole & Enchiladas	Ravioli	Lamb Chops
Horchata	Burrata	Red Lentil Soup
	Cioppino	Baklava

Adding a cuisine is as simple as editing dishes.js.

Quick start

# 1. Clone and set up keys
git clone https://github.com/selinayfilizp/sf-eats.git
cd sf-eats
cp .env.example .env
# Edit .env — add your GOOGLE_PLACES_API_KEY (required) and SERPAPI_KEY (optional)

# 2. Try it offline first (no keys needed)
node generate-data.js --dry-run
open index.html

# 3. Generate live data
npm install           # installs convex (only dependency)
node generate-data.js
open index.html

Requires Node 18+ (uses global fetch).

API keys you need

Key	Required	Free tier	What it does
GOOGLE_PLACES_API_KEY	Yes	$200/mo credit	Finds restaurants, ratings, locations
SERPAPI_KEY	No	100 searches/mo	Pulls full review lists for hype/love scoring

Without a SerpAPI key, spots still show up with ratings and locations — they just won't have hype/love scores (shown as "not yet rated").

Community recs

The app includes a Community Picks section below the ranked spots for each dish. Anyone can submit their recommendation for where to get a dish, with a note about why it's the best. Submissions are stored in Convex and shared across all users in real time.

Project structure

index.html      # The entire front-end (single file)
generate-data.js      # CLI data generator (Google Places + reviews)
dishes.js             # Cuisine/dish catalog — edit to add dishes
reviews-provider.js   # SerpAPI/Outscraper review fetching + scoring
sentiment.js          # Lightweight dish-sentence sentiment analysis
verified-dishes.js    # Manual override layer for hype/love scores
test-reviews.js       # Quick single-place sanity check
convex/               # Convex backend for community recommendations
  schema.ts           #   Database schema (recs table)
  recs.ts             #   Query + mutation functions

Contributing

Want to add a cuisine or improve the scoring? PRs welcome. The main things to know:

• sf-food-data.js is generated — don't edit it by hand
• Never commit .env or API keys
• Keep it dependency-free and single-file-per-concern

License

MIT