Back to Home
Retrieval-Augmented Generation (RAG)

Re-ranking

Re-ranking means checking retrieved chunks again and putting the most useful ones at the top.

This is a simplified learning demo. Real systems may use special re-ranking models.

Interactive Playground

Retrieved Chunks

70
Live Visualization
📄 Showing original retrieval order

Retrieved Order

Re-ranked Order

Statistics

5
Total Retrieved
Top Before
Top After
0
Chunks Moved
70%
Re-ranking Strength

How It Works

📥
Retrieved
Chunks
🧮
Re-score
🔀
Reorder
🤖
Better Context
for AI
Retrieval order isn't always the final order. Re-ranking gets a second, closer look at each chunk.
Set Strength to 0 to keep the original order, or 100 to fully trust the re-ranker.
Try the random examples — chunks that barely relate to a question drop down, on-topic ones rise to the top.
💡
Key Takeaway

Re-ranking helps the AI receive the best chunks first, improving the final answer.