Vector Databases: Embeddings, Semantic Search, and Hybrid Retrieval - Alexey Grigorev
In this workshop, Alexey Grigorev, founder of DataTalks.Club, dives deep into the technical shift from lexical to semantic retrieval in modern AI applications. We explore the evolution of RAG pipelines, specifically moving from basic text search to advanced vector search architectures and production-ready deployments.
Links:
LLM Zoomcamp: https://github.com/DataTalksClub/llm-...
Materials: https://github.com/DataTalksClub/llm-...
Guardrails workshop: https://aishippinglabs.com/workshops/...
You’ll learn about:
Why meaning-based search beats keyword matching
How to shift from lexical to semantic retrieval in RAG pipelines
How to implement vector search with PGVector and SQLite
How to use ONNX to make AI models 33x lighter and faster
How to manage large datasets with Approximate Nearest Neighbors (ANN)
How to automatically filter irrelevant or out-of-scope queries
A clear guide to Embeddings, Cosine Similarity, and Scoring
TIMECODES:
0:00 Evolution from Text Search to Vector Search
7:33 RAG Architecture and Prompt Engineering with Context
11:26 Environment Setup for Sentence Transformers and PyTorch
16:45 Fundamentals of Word and Sentence Embeddings
20:09 Utilizing Sentence-BERT for Semantic Similarity
25:16 Mathematics of Vector Similarity and Cosine Similarity
28:17 FAQ Dataset Embedding and Batch Processing Workflows
33:53 High-Performance Matrix Multiplication for Similarity Scoring
38:18 Implementing Top-K Search and Argmax using NumPy
43:23 MinSearch Integration and Metadata Filtering Strategies
48:54 Hybrid Search Advantages and Lexical-Semantic Integration
52:30 Refactoring RAG Pipelines for Vector Search Swapping
57:59 Benchmarking and Justification for Vector Search Deployment
1:02:47 Approximate Nearest Neighbors vs Exact Search Performance
1:07:01 Decoupling Data Ingestion from Production Deployment
1:12:44 Vector Index Implementation with SQLite Search
1:18:48 Production Vector Search using PGVector and Postgres
1:23:44 LLM Guardrails and Filtering Irrelevant User Queries
1:35:13 ONNX Model Optimization for Lightweight Production AI
Connect with DataTalks.Club:
Join the community - https://datatalks.club/slack.html
Subscribe to our Google calendar to have all our events in your calendar - https://calendar.google.com/calendar/...
Check other upcoming events - https://lu.ma/dtc-events
GitHub: https://github.com/DataTalksClub
LinkedIn - / datatalks-club
Twitter - / datatalksclub
Website - https://datatalks.club/
Connect with Alexey
Twitter - / al_grigor
Linkedin - / agrigorev
Check our free online courses:
ML Engineering course - http://mlzoomcamp.com
Data Engineering course - https://github.com/DataTalksClub/data...
MLOps course - https://github.com/DataTalksClub/mlop...
LLM course - https://github.com/DataTalksClub/llm-...
Open-source LLM course: https://github.com/DataTalksClub/open...
AI Dev Tools course: https://github.com/DataTalksClub/ai-d...
👉🏼 Read about all our courses in one place - https://datatalks.club/blog/guide-to-...
👋🏼 Support/inquiries
If you want to support our community, use this link - https://github.com/sponsors/alexeygri...
If you’re a company, reach us at [email protected]
ag #vectorsearch #semanticsearch #llmzoomcamp #aiengineering #nlp #embeddings #pgvector #postgres #sqlite #onnx #aioptimization #machinelearning #generativeai #datatalksclub #python #searchinfrastructure #cosinesimilarity #hybridsearch #semanticretrieval