Course Objective & Scope
Vector search sits at the core of semantic search, RAG pipelines, recommendation engines and multimodal retrieval. Yet at millions of high-dimensional vectors, misconfigured index parameters, naive metadata filters or unquantized float32 vectors quickly lead to either runaway RAM costs or poor search recall.
Rather than tying to a single vendor, this course focuses on the core architecture and algorithms behind vector databases. It covers dense and sparse (SPLADE/BM25) representations, Cosine/Dot Product/L2 metrics, HNSW, IVF and DiskANN indexing, memory reduction with Product, Scalar and Binary Quantization, pre/post/single-stage metadata filtering, multi-tenancy, and Recall@K vs QPS benchmarking.
Who Should Attend?
- AI and backend engineers building semantic search, RAG and recommendation systems
- Data and solution architects planning enterprise vector search capacity and topology
- Search and data science teams tuning retrieval quality (Recall/NDCG) and query latency
- Prerequisite: basic Python or backend development experience and core database familiarity
Key Takeaways
- Match Cosine, Inner Product or Euclidean (L2) distance metrics to how the embedding model was trained
- Tune HNSW (
M,efConstruction,efSearch) and IVF (nlist,nprobe) parameters across the Recall–Latency–RAM frontier - Cut memory footprint by 75–95% using Scalar, Product (PQ) and Binary Quantization with rescoring
- Prevent filtered-ANN recall drops caused by restrictive metadata predicates
- Combine dense semantic and sparse lexical search with Reciprocal Rank Fusion (RRF) and benchmark Recall@K
Curriculum
Embedding Representations, Dimensionality and Distance Metrics
- Dense, sparse (SPLADE), multi-vector (ColBERT late interaction) and Matryoshka (MRL) embeddings
- Cosine Similarity, Inner Product (Dot Product) and Euclidean (L2) distance, plus vector normalisation
- The curse of dimensionality and when Exact Search (Flat / Brute-Force k-NN) stops scaling
Approximate Nearest Neighbor (ANN) Indexing Algorithms
- Cluster-based indexes: Inverted File (IVF), Voronoi cells and balancing
nlistvsnprobe - Graph-based indexes: HNSW multi-layer graphs and tuning
M,efConstructionandefSearch - Disk-resident indexes: Vamana / DiskANN architecture for billion-scale SSD-backed search
Memory Sizing and Vector Quantization Techniques
- Calculating actual RAM usage for float32 vectors plus HNSW graph link overhead
- Scalar Quantization (FP16 / INT8) and Product Quantization (PQ subspace codebooks)
- Binary Quantization (1-bit) combined with oversampling and full-precision rescoring
Metadata Filtering, Hybrid Search and Multi-Tenancy
- The filtered-ANN problem: comparing Pre-filtering, Post-filtering and Single-Stage (payload-aware) traversal
- Combining Dense + Sparse/BM25 retrieval with Reciprocal Rank Fusion (RRF) and Cross-Encoder reranking
- Multi-tenant isolation: per-tenant namespaces/shards vs shared-index metadata filtering
Benchmarking, Mutation Handling and Production Operations
- Measuring Recall@K, NDCG@K, QPS and p95/p99 latency across index configurations
- How deletes, updates (tombstones) and background segment compaction affect HNSW graphs
Other Courses in This Category
-
Hands-on engineering across the four leading vector platforms: indexing, hybrid search, multi-tenant isolation and architectural selection with PostgreSQL pgvector, Milvus, Weaviate and Pinecone.
- Live Course
- 5 modules
-
Cost-Based Optimizer internals, reading Execution Plans, SARGability, B-Tree/Covering/Filtered index design, join algorithms, parameter sniffing, lock/isolation analysis and partitioning — for developers and DBAs.
- Live Course
- 5 modules
-
PL/SQL block structure, cursors, modular Package design with procedures and functions, enterprise exception handling, triggers, dynamic SQL and high-volume processing with FORALL and BULK COLLECT.
- Live Course
- 5 modules
-
Conceptual/logical/physical data modelling, 1NF–3NF/BCNF normalization rules, preventing data anomalies, controlled denormalization, natural vs surrogate keys, constraints and zero-downtime schema evolution.
- Live Course
- 5 modules
-
SQL Querying Fundamentals
Foundational SQL Server course covering relational database concepts, data modeling, normalization, indexing, database objects (Tables, Views, Stored Procedures), and JOIN queries.
- Live Course
- 8 modules
-
SQL Server Querying
Comprehensive SQL Server and T-SQL querying course covering SSMS, SELECT workflows, multi-table JOINs (INNER, OUTER, CROSS, SELF), filtering and pagination, data types, and built-in Scalar, Aggregate, and Window functions.
- Live Course
- 7 modules
Get in touch
