Chunking and freshness

Chunking and document freshness

• Chunking Strategies and Their Effect on Retrieval
• What a Model Sees: Assembling Context from Chunks
• Building a Chunking Pipeline That Survives Document Updates
• Choosing a Chunk Size for Tables and Code
• Choosing a Chunk Size That Balances Recall and Token Cost
• Chunking Decisions and the Queries They Break
• Deduplicating Chunks Before They Reach the Model
• Keeping Chunks Fresh When Source Documents Change
• All of Chunking and freshness

Embeddings in Postgres

Storing embeddings in Postgres

• How to Build Embeddings and Vector Search
• Dual Writing to Two Vector Stores and Cutting Over
• Embedding Version as a Column, Not a Config Value
• One Embedding Service for Several Products
• Storing a Model Version Beside a Vector: Schema and Backfill
• Storing Chunks and Metadata in Postgres with pgvector
• When to Denormalize for Retrieval: Columns vs Joins

Vector index tuning

Vector index tuning and rebuilds

• Choosing Between HNSW and IVFFlat After Your Data Grows
• pgvector at Scale: Vacuum, Reindexing and Index Rebuilds
• Tuning HNSW Parameters Against a Recall Target You Can Measure
• A Dual-Write Table So a Reindex Can Be Resumed
• A Second Vector Index Beside the First: Build, Validate, Cut Over
• Blue-Green Rebuilds for Vector Indexes: A Step-by-Step Cutover
• Choosing a pgvector Index Build for a Table You Cannot Lock
• Detecting Index Bloat with pg_stat and Vector Index Stats
• All of Vector index tuning

Choosing embedding models

Choosing embedding models

• Choosing an Embedding Model by Dimensions, Cost and Fit
• Deciding Whether a New Embedding Model Earns a Full Reindex
• Switching an Embedding Provider Behind a Gateway Route
• A Budget for Embed Calls: Cost Per Query and Per Backfill Job
• Choosing an Embedding Model: Dimensions, Cost, Fit
• Cutting Vector Dimensions Without Losing Recall
• Embeddings for Code, Tables and Images
• Estimating a Reindex Bill from a Token Count
• All of Choosing embedding models

Hybrid search and reranking

Hybrid search and reranking

• Rerankers in Practice: Cross-Encoders, Latency and Cost
• A Reranking Microservice: Batching, Timeouts and Fallbacks
• Hybrid Search: BM25 and Vectors Without Doubling Latency
• Merging Keyword and Vector Results with Reciprocal Rank Fusion
• Merging Keyword and Vector Results: Tuning Alpha on Your Test Set
• When to Add a Reranker After the Vector Search Returns
• When to Add a Reranker: Cost and Latency Tradeoffs

Retrieval quality

Retrieval quality evaluation

• A Retrieval Test Set for Embedding Model Swaps
• Measuring Retrieval Quality: Recall and Precision
• Writing a Regression Query for Every Retrieval Bug
• A Query Rewrite Step: When the Model Paraphrases First
• A Relevance Floor on Context: Setting It from the Judged Set
• Connecting Retrieval Scores to Support Outcomes
• Generating Realistic Query Sets for Retrieval Tests
• Measuring Recall@k and Precision on a Golden Set
• All of Retrieval quality