Lab 6: Hybrid Search – Combining Semantic & Keyword Matching¶

Level: Intermediate-Advanced | Duration: 2.5 hours

Objective¶

Implement production-ready hybrid search combining dense (semantic) and sparse (keyword) retrieval.

What You'll Learn¶

• Implement BM25 sparse retrieval algorithm
• Understand TF-IDF scoring
• Implement Reciprocal Rank Fusion (RRF)
• Combine multiple ranking signals
• Compare hybrid vs pure semantic search
• Optimize and tune ranking parameters
• Evaluate with relevant metrics

Hybrid Search Flow¶

Query
  ↓
├─→ Dense Retrieval (Semantic) → Top-k results + scores
├─→ Sparse Retrieval (BM25)     → Top-k results + scores
  ↓
Combine Rankings (RRF)
  ↓
Final Ranked List
  ↓
Return to User

When to Use Hybrid¶

• ✅ Exact matches important (IDs, codes)
• ✅ Mix of semantic + keyword queries
• ✅ Domain-specific terminology
• ✅ Production systems needing robustness
• ✅ When either semantic or keyword alone fails