[DOC] Deep walkthrough: Hybrid retrieval + reranking + evaluation

[tazarov]

Why

Teams need a complete retrieval quality pipeline with measurable outcomes, not just isolated algorithm notes.

Proposal

Create an end-to-end walkthrough for hybrid retrieval with reranking and evaluation.

Scope

• Ingest a realistic corpus.
• Implement dense retrieval + lexical retrieval (BM25) + fusion (RRF).
• Add reranking stage.
• Evaluate with a labeled set and report metrics before/after each stage.

Deliverables

• Runnable pipeline example.
• Configurable chunking/indexing parameters.
• Metrics table (baseline vs hybrid vs hybrid+rerank).
• Guidance on interpreting quality/latency/cost tradeoffs.

Acceptance Criteria

• The walkthrough produces reproducible evaluation metrics.
• At least one metric improves from baseline to hybrid+rerank.
• The guide includes guardrails for common tuning mistakes.