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
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
Chunks
→
🧮
Re-score
→
🔀
Reorder
→
🤖
Better Context
for AI
for AI
🎓
Tips
3 tips
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.