Chunking Strategies for Long-Form Web Documents
Chunking strategy matters more than embedding model choice for retrieval quality.
Marcus Oyelaran
Section
RAG Architecture Patterns
10 stories in RAG Architecture Patterns.
Graph-Based Retrieval Augmentation With Knowledge Graphs
Graph-based retrieval handles multi-hop questions vector search alone cannot answer correctly.
Jailen Rafferty
RAG vs. Long-Context LLM Windows for Document QA
Four variables determine whether RAG or long-context windows will cost less and work better.
Marcus Oyelaran
Evaluating RAG Pipeline Quality With RAGAS
RAGAS separates retrieval failures from generation failures without labeled answer keys.
Jailen Rafferty
Multi-Tenant RAG Index Isolation Patterns
Separate indexes enforce isolation at the storage layer, not the prompt.
Jailen Rafferty
Reranking Models as a RAG Retrieval Layer
Cross-encoders beat single-vector retrieval by rescoring candidates jointly instead of in isolation.
Marcus Oyelaran
Hybrid Search With Dense and Sparse Vectors in RAG
Combining dense vectors with BM25 retrieval fixes the two ways RAG systems fail most often.
Priya Subramaniam
Metadata Filtering to Reduce Retrieval Noise in RAG
Filtering retrieved documents by metadata slashes hallucination rates by half or more.
Marcus Oyelaran
Index Freshness Strategies for Live-Web RAG
Decide staleness tolerance per topic, not per pipeline.
Jailen Rafferty
Chunk Size Tradeoffs in RAG Retrieval
Smaller chunks excel at factoid queries while larger ones catch thematic relationships.
Marcus Oyelaran