Abalone RAG Demo

This RAG system uses SBERT for initial retrieval and a Cross Encoder for re-ranking and highlighting.

Sentence embeddings are computed and indexed using FAISS.

For generation, you can choose between:

• FLAN-T5 — Fast and reliable, the baseline experience.
• Finetuned TinyLlama — Slower, but more expressive.
• No Generation — Only retrieve and highlight relevant context without generating a response. Explore the retrieval quality.