← All Projects
02 In Progress

AI Coffee Pairing

Using LLMs to suggest coffee origins and brewing methods based on your flavor profile and preferences.

GitHub ↗

The problem

Walk into any specialty shop and you'll find a wall of origins, processes, and roast dates — but almost no way to translate "I like bright, juicy, low-acid coffee" into an actual bag to buy or a brew method to use. Tasting notes on packaging are inconsistent, and most recommendation tools just upsell whatever's in stock.

The solution

AI Coffee Pairing takes a plain-language flavor description — or a set of cupping scores straight from the Cupping Tool — and matches it against a structured dataset of origin, process, and roast characteristics using an LLM prompted with SCA flavor wheel vocabulary. It returns a ranked shortlist of origins and a recommended brew method, with the matching water profile from Water Crafter already attached.

Tech stack

SvelteKitOpenAI APIQdrantTypeScript

Timeline

Dec 2025

First prototype

Built a prompt-based recommender against a hand-labeled set of 40 origins using GPT-4.

Feb 2026

Cupping data pipeline

Connected Cupping Tool exports as a live reference dataset instead of static labels.

Apr 2026

Vector search

Moved matching to Qdrant embeddings so recommendations scale past a few dozen origins.

Jun 2026

Private beta (current)

Testing pairing accuracy with a small group of newsletter subscribers before a public launch.

Other projects