Embeddings & Vector Search
Understand how text becomes vectors and how to search them: similarity measures, learned embeddings, exact and approximate nearest neighbours, quantisation, hybrid search, evaluation and production, with every experiment run for real.
Syllabus
What Embeddings Are
- From Words to Coordinates
- Where Embeddings Come From
- Similarity Measures: Dot, Cosine and Euclidean
- Dimensions and the Curse of Dimensionality
Creating and Choosing Embeddings
- A Learned Embedding You Can Inspect
- Keyword Search vs Semantic Search
- Choosing an Embedding Model
- Preparing Text: Chunking, Truncation, Query vs Document
Searching Vectors: Exact and Approximate
- Exact k-Nearest Neighbours
- Approximate Search: IVF and HNSW
- Compression: Scalar and Product Quantisation
- Choosing an Index
Beyond Plain Nearest Neighbours
- Metadata Filtering
- Hybrid Search and Reranking
- Clustering, Topics and De-duplication
- Multi-Vector, Multilingual and Multimodal Embeddings
Evaluating and Improving Retrieval
- Recall@k, MRR and nDCG on a Golden Set
- Error Analysis and Improving Quality
- Model Changes, Re-Indexing and Drift
Running Vector Search in Production
- Where Vectors Live
- Updates, Deletes and Freshness
- Scaling, Cost and Latency
- Security and Privacy of Embeddings
Applications of Embeddings
- Semantic Search and Recommendations
- Classification, Routing and Few-Shot Labelling
- Embeddings for RAG and Agent Memory