MemeMeow

Why we included this project

MemeMeow searches memes by meaning instead of filename. You type a description of the scene you want, and an embedding model matches it to the meme whose caption fits best, the same retrieval pattern behind larger RAG systems. It runs locally with a Streamlit interface and also exposes a simple HTTP API, so you can wire the same search into a bot, an iOS shortcut, or a script without deploying anything. The resource-pack model is what makes it interesting for teams: images are labeled, optionally with a VLM, bundled into shareable packs, and shared through the community, which shows how a small project organizes a growing, community-contributed dataset. It's a compact, low-cost way to see embeddings, vector matching, and image labeling working end to end.