Sommelier AI
Wine recommendations that don't suck
The backstory
My entire family works in gastronomie—hotels and restaurants across North America and Switzerland. They've owned and run these places for years. This is the wine assistant they'd actually want to use.
What it does
- •Scraped tens of thousands of wines from a Swiss retailer
- •Sophisticated scraper: SOCK5 IP rotation, user agent switching, the whole stealth toolkit
- •Vectorized everything with Supabase for RAG
- •Interactive chat: "yo I'm cooking seafood pasta" → get wine pairing recommendations seamlessly
- •Full-stack app teaching users about wine without the pretentious sommelier vibe
The scraping story
Scraping tens of thousands of wine products isn't just a for-loop. We're talking SOCK5 proxies, rotating IPs, user agents that make you look like grandma's Firefox— whatever it takes to stay undetected. Think Ocean's Eleven but for wine data.
How the pairing works
You chat naturally: "I'm thinking of cooking something with seafood, perhaps seafood pasta." The RAG system pulls relevant wines from the vectorized database, and the LLM explains why they work. No wine snobbery, just useful recommendations.
Completely self-hosted and containerized. The scraper runs on schedule, the vector database is ours, and the frontend talks to our own backend. No third-party dependencies for user data.