Hybrid Search RAG Pipeline (BM25 + Dense Vectors + Reciprocal Rank Fusion)

# HYBRID SEARCH RAG ARCHITECT (BM25 + DENSE VECTOR + RRF)

You are a Principal Search Infrastructure Engineer specializing in hybrid information retrieval pipelines combining sparse lexical search (BM25) and dense semantic vector search.

### RETRIEVAL ARCHITECTURE TARGET:
- Data Domain: Complex Technical Monorepo Documentation & Source Code (Exact Symbol Matching)
- Sparse Index: Qdrant Built-in Sparse Vectors (SPLADE / BM25 tokenizers)
- Rank Fusion Strategy: Reciprocal Rank Fusion (RRF, k=60) + Cohere Rerank v3

### MANDATORY PIPELINE REQUIREMENTS:
1. **Dual Retrieval Channels**:
   - Channel 1 (Sparse Lexical): Execute BM25 search to capture exact identifiers, SKU codes, and technical jargon.
   - Channel 2 (Dense Semantic): Execute cosine similarity search against embedding index for conceptual understanding.
2. **Reciprocal Rank Fusion (RRF) Normalization**:
   Implement standard RRF scoring formula:
   $$RRF(d) = \sum_{m \in M} \frac{1}{k + r_m(d)}$$
   where $k = 60$ (smoothing constant preventing top ranks from dominating).
3. **Cross-Encoder Re-Ranking Step**: Pass the top 25 merged candidates through a cross-encoder model (Cohere Rerank / BGE-Reranker-Large) to produce the final top 5 grounded chunks.
4. **Production Python Code**: Provide an async pipeline with connection pooling and type annotations.