RAG and vector search

Rank documents by dot product, cosine and L2 distance and explain the role of normalization and model choice.

1. R1.1Query and document encoders
2. R1.2Dot product
3. R1.3Cosine similarity
4. R1.4L2 distance
5. R1.5Dimensions and storage
6. R1.6Retrieval evaluation

Implement an exact baseline, compute recall at k and explain approximate search and high-dimensional difficulty.

1. R2.1Exact search
2. R2.2Top-k selection
3. R2.3Recall at k
4. R2.4High dimensions
5. R2.5Trees and hashing
6. R2.6Approximation tradeoffs

Trace IVF and HNSW search and explain how candidate exploration trades latency for recall.

1. R3.1Inverted files
2. R3.2Probing and candidate counts
3. R3.3Navigable graphs
4. R3.4HNSW layers
5. R3.5Disk-based search
6. R3.6Tuning and updates

Calculate vector-code storage and distinguish scalar, product and binary quantization from reranking.

1. R4.1Scalar quantization
2. R4.2Product quantization
3. R4.3Codebooks and lookup tables
4. R4.4Binary representations
5. R4.5Reranking
6. R4.6Quality and memory budgets

Design a retrieval record and explain filtering, indexing, partitioning, updates and consistency.

1. R5.1Records and identity
2. R5.2Exact versus approximate lookup
3. R5.3Filtering
4. R5.4Sharding and replicas
5. R5.5Updates and deletes
6. R5.6Hybrid storage and operations

Create source-traceable chunks, calculate overlap and explain ingestion choices that preserve answer-bearing context.

1. R6.1Parsing and structure
2. R6.2Fixed-size chunks
3. R6.3Overlap
4. R6.4Semantic boundaries
5. R6.5Metadata and contextual headers
6. R6.6Incremental ingestion

Fuse rankings, rerank evidence and evaluate retrieval separately from grounded answer generation.

1. R7.1Lexical and dense search
2. R7.2Hybrid fusion
3. R7.3Rerankers
4. R7.4Query transformation
5. R7.5Grounding and citations
6. R7.6Retrieval metrics