https://github.com/tgenaitay/wineismine
Query tastes and fetch a selection of great wines from portfolio
https://github.com/tgenaitay/wineismine
Last synced: 3 months ago
JSON representation
Query tastes and fetch a selection of great wines from portfolio
- Host: GitHub
- URL: https://github.com/tgenaitay/wineismine
- Owner: tgenaitay
- Created: 2025-03-17T09:04:49.000Z (3 months ago)
- Default Branch: main
- Last Pushed: 2025-03-18T01:02:13.000Z (3 months ago)
- Last Synced: 2025-03-18T02:22:17.884Z (3 months ago)
- Language: HTML
- Size: 35.2 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Wine is Mine 🇫🇷🍷
A web application to query tastes from wine enthusiasts, and fetch a selection of great wines from our catalog.
Currently using Together AI's free endpoint for consuming Llama 3.3.
## Features
- Server-side processing with Node.js
- Static file serving for frontend assets
- LLM integration for intelligent transformation of user input into filters
- Function call to Supabase for querying best wines
- Scoring based on user's preferences and WIM notes
- Data storage all in Supabase
## Prerequisites
- Node.js
- npm
- Supabase
- Wine data :)
## Setup
1. Clone the repository
2. Install dependencies:
```bash
cd server
npm install
```
3. Create a `.env` file in the root directory with required environment variables
```
TOGETHER_API_KEY=xx
SUPABASE_URL=xx
SUPABASE_KEY=xx
```
## Development
To run the development server (nodemon used for hot reload):
```bash
cd server
npm start
```
## Deployment
This project is configured for deployment on Vercel. The `vercel.json` configuration handles:
- Node.js server deployment
- Static file serving
- Route configurations
To deploy:
1. Install Vercel CLI
2. Install environment variables in Vercel settings
3. Run `vercel --prod` in the project root
## Demo
https://wine-is-mine.vercel.app/
All feedback welcome.
## Known issues
- Llama 3.3 randomly outputs a malformed native function call in its content instead of a regular tool call. Happens 1 out of 5 times. Moving away from Llama to e.g Gemini should make it more stable.
## Wine Selection Logic 🍇
Our selection process works in 3 clear phases:
1. **Initial filtering**
We first look for up to 20 wines that may match :
- Preferred regions (e.g. Burgundy/Provence)
- Specific appellations (e.g. Meursault)
- Favorite domains/châteaux
- Price range preferences
If the initial search resulted to less than 10 wines, or if the connoisseurs let us broaden their horizons, we expand the search without filters - except for price.
2. **Scoring**
Each wine earns points based on:
- **Color priority**: 4x points for 1st choice color, 3x for 2nd, etc, with a factor of 10
- **Taste match**: Matches wine characteristics to user preferences
*(e.g. "Fruity reds" get boosted if user prefers fruity flavors)*
- **Quality score**: Wine is Mine internal rating (note_wim) gives bonus points, with a factor of 5
3. **Final selection**
- up to 20 potential matches are identified
- These are re-ranked by total score
- Final selection shows the 10 highest-scoring wines
This ensures recommendations balance personal preferences with best in class wines.