Reranking: Cross-Encoder vs Bi-Encoder - Interactive Visualization

Bi-encoders encode query and document separately and compare via cosine similarity - fast but imprecise. Cross-encoders jointly encode both and produce a relevance score - slower but much more accurate. Reranking applies a cross-encoder to the top-K bi-encoder results, getting the best of both. This demo shows how reranking reorders results and improves Recall@1.

• Bi-encoder stage: query and document embeddings computed independently at millisecond speed
• Cross-encoder reranking: joint encoding of query+document pairs produces finer relevance scores
• Recall@1 comparison: see the jump in top result accuracy after applying a reranker
• Latency breakdown: retrieval vs reranking time shown separately to quantify the cost

Part of the EngineersOfAI Interactive 3D - free interactive visualizations covering every major concept in machine learning and AI engineering. Hover any element for a plain-English explanation. No code required.