Mastering Vector Databases & Embedding Models in 2025

Name: Mastering Vector Databases & Embedding Models in 2025
Rating: 4.5 (28 reviews)

Learn embeddings, similarity search, HNSW, IVF, semantic search, RAG, and recommender systems with hands-on examples.

Highest Rated

Created byTensor Teach

Last updated 8/2025

English

English [Auto],

What you'll learn

Explain what embeddings are and how they enable similarity search.
Learn how to choose and fine-tune embedding models for custom applications.
Learn how vector databases work in terms of indexing & retrieval.
Familiarize yourself with the vector database landscape and different applications.

Course content

4 sections • 22 lectures • 1h 58m total length

What Are Embeddings?4:16
Explore how embeddings turn unstructured data into high-dimensional numerical vectors, enabling vector databases to capture semantic meaning and support similarity search, clustering, and retrieval augmented generation in machine learning tasks.
Section 1 Notebook0:03
Creating & Visualizing Embeddings with Sentence Transformers5:50
Learn to create and visualize text embeddings with sentence transformers, encode batches, and plot semantic clusters in a vector space, paving the way for efficient similarity in vector databases.
Understanding Vector Similarity Metrics6:22
Explore cosine similarity, dot product, and Euclidean distance for sentence transformer embeddings. Learn why dot product is faster when vectors are normalized, with numpy examples for production use.
Creating a Mini-Search Engine Using Embeddings2:45
Demonstrates building a tiny search engine with embeddings, using semantic search to rank sentence-level results by query similarity and top-k selection.

Understanding Embedding Models: Tokenizer & Architecture13:03
Explore how embedding models are built from encoder-only architectures like Bert and Roberta, trained with positive and negative pairs, and produced via tokenization, pooling, and normalization to generate single vectors.
Evaluating and Selecting Text Embedding Models6:38
Learn how to evaluate and select text embedding models using the massive text embedding benchmark (MTM) on Huggingface, with multilingual coverage, retrieval focus, and domain-specific evaluation.
Multimodal Models - A Brief Overview3:57
Explore multimodal models fusing text and image inputs, creating shared embeddings and similarity between text and images, using dual, fusion, and hybrid architectures trained with contrastive learning.
Working with Multimodal Models in Transformers7:05
Explore using the clip model from Transformers to process images and text with separate encoders and tokenizers, then compute embeddings and cosine similarity for multimodal retrieval.
Fine-Tuning Embedding Models with Contrastive Loss12:39
Fine-tune embedding models with contrastive loss on domain data to keep them relevant. Train with triplets and hard negatives to improve distinction between related and unrelated queries.

Introduction to Vector Database Indexing & Retrieval Strategies5:22
Explore how vector databases index and retrieve data using approximate nearest neighbors, including hnsw (hierarchical navigable small worlds) and the inverted file system for fast, scalable search.
HNSW - Indexing & Retrieval Explained6:57
Explore how HNSW builds vector databases with a multi-layer graph of nodes and edges, detailing insertion to bottom-to-top layers, and top-down retrieval to return the k closest vectors.
Section 3 Notebook0:04
HNSW Implementation in FAISS6:55
Implement HNSW in FAISS to build a hash index, encode facts and questions into 384-dimensional embeddings, and retrieve top-k results for a question-answering system.
IVF - Indexing & Retrieval Explained6:10
Explore how inverted file indexing clusters embeddings with k-means into centroid groups, stores vector ids in posting lists, and retrieves by probing closest clusters to balance speed and recall.
IVF Implementation in FAISS2:29
Learn how to implement an IVF index in FAISS, with a 384-dimensional embedding, five clusters, and two-cluster probing. Train with KMeans, add to index, and tune nprobe and top-k results.

Overview of Vector Database Landscape + Pinecone Introduction5:25
Explore the vector database landscape, from pure and supported databases to open, closed, and hybrid options, and get introduced to Pinecone for vector search and embedding models.
Section 4 Notebook0:01
Semantic Search with Pinecone8:03
Discover semantic search as the backbone of vector databases through embedding models, building a Pinecone index, chunking data, and running semantic queries with embeddings.
Retrieval Augmented Generation (RAG) with Pinecone6:03
Explore retrieval augmented generation (rag) with Pinecone to improve LLMs' responses by retrieving relevant context via embedding models and vector search.
Recommender Systems with Pinecone4:58
Explore recommender systems with vector databases and embedding models, using user and content embeddings to retrieve semantically aligned news articles for a given profile with Pinecone.
Bonus Lecture - Free AI Research Newsletter3:18
Discover a free, weekly AI research newsletter that distills arXiv papers on multimodal LLMs, embodied agents, robotics, and quantization, highlighting the research question, methods, results, and key takeaways.

Requirements

Basic Python knowledge recommended, but step-by-step coding lessons are provided.
No prior experience with embeddings, vector databases, or similarity search is required.

Description

Embeddings and vector databases are the foundation of many modern AI applications — from semantic search to retrieval-augmented generation (RAG) and personalized recommendations. This course takes you from the core concepts to production-ready solutions, following a structured, project-based approach.

In Section 1, you’ll build deep intuition about embeddings: what they are, how they are produced with Sentence Transformers, and how similarity metrics like cosine, Euclidean, and dot product work. You’ll then apply these concepts to build a mini search engine.

In Section 2, you’ll learn how to choose and customize embedding models. We’ll cover how embedding models are formed, how to evaluate them using the MTEB benchmark, and how to use multimodal embeddings. You’ll then explore fine-tuning with contrastive loss.

In Section 3, we go under the hood of vector databases. You’ll learn the theory behind indexing methods like HNSW and IVF through clear visual explanations, followed by coding demos showing them in action.

In Section 4, we turn theory into practice. You’ll explore the vector database landscape, implement semantic search and dense retrieval, integrate embeddings into RAG pipelines, and build recommender systems using Pinecone — all with reproducible Python notebooks.

By the end of this course, you’ll have both the conceptual understanding and the hands-on skills to confidently build and deploy AI applications powered by embeddings and vector databases.

Who this course is for:

Data scientists, ML engineers, and software developers exploring vector search.
AI practitioners and enthusiasts who want to apply embeddings in real-world projects.
Professionals interested in semantic search, RAG, and recommender systems.

Mastering Vector Databases & Embedding Models in 2025

What you'll learn

Explore related topics

Course content

Foundations of Embeddings & Similarity5 lectures • 19min

Choosing and Using Embedding Models5 lectures • 43min

How Vector Databases Work6 lectures • 28min

Vector Databases: Landscape & Applications6 lectures • 28min

Requirements

Description

Who this course is for: