Vector Databases Fundamentals to Production [2026 Edition]

Click the resources area to access PDFs and slides, or right-click the lecture to open the source code link. Retrieve the actual code for the course from that link.

Explore how vector databases encode information as vectors in a multi-dimensional space to enable efficient similarity search, compare them with traditional databases, and grasp the role of dimensionality.

Learn why vector databases excel at unstructured data by transforming items into vectors for fast similarity search and nuanced feature-based matching.

Explore how vector databases convert complex data into numerical vectors for quick similarity search, real-time interactions with machine learning and large language models, and scalable performance via indexing.

contrast traditional relational databases with vector databases, outlining SQL, structured data, schema constraints, ACID, indexing, and full table scans, versus vector databases' strength in unstructured data and high-dimensional similarity search.

Transform raw unstructured data into embeddings to create semantic vector representations, and use vector databases with nearest-neighbor indexing and distance metrics to find similar items.

Transform unstructured data into embeddings and store them in a vector database; query via embedding to retrieve similar entries via an index and feed them to a large language model.

See how embeddings turn text into meaning-bearing vectors and power the document-to-answer workflow in vector databases. Distinguish embedding models from chat models and compare dimensions, trade-offs, and search speed.

Index and chunk documents, embed each chunk with a consistent embedding model to build a vector store, then query, search, retrieve, augment, and answer with a chat model.

Understand how embeddings differ from vectors: embeddings are learned vectors that encode semantic similarities and metadata, while vectors are generic mathematical representations in machine learning and artificial intelligence.

Explore the wide range of vector databases use cases, from image retrieval and real-time similarity search in e-commerce to personalized music recommendations, NLP-driven chatbots, fraud detection, and bioinformatics.

Compare traditional and vector databases, noting limitations and contrasts, and learn about embeddings and vector databases, including what they are, how they work, advantages, and use cases.

Explore the top five vector databases—pinecone, milvus, face, chroma, and annoy—and compare strengths like real-time indexing, open-source options, and fast similarity search.

Understand how large language models train on vast text data to generate human-like text, using unsupervised learning, supervised learning, and encoder-decoder transformers with self-attention to capture context.

Paolo invites you to rate the course and share feedback. Post questions in the discussion board and help answer them to foster a collaborative learning community.

Set up your development environment and prepare the tools to build vector databases, including VS Code, Python, and an OpenAI API key, before starting hands-on construction from scratch.

Install Visual Studio Code and set up Python on your computer, then create and securely store an OpenAI API key to enable vector database work.

Understand the chroma database workflow for vector databases, embeddings, and documents. See how queries convert to embeddings, drive semantic search, and feed a large language model to generate answers.

Learn to set up a chroma vector database in Python, create or get a collection, upsert documents, and run queries to retrieve closest matches with similarity distances.

Learn to loop through similarity search results in Chroma DB, extract document IDs and distances, and display the most similar documents for a given query.

Learn how to create embeddings with chroma db using the default embedding function and attach it to a collection, then explore integrating with OpenAI and other vendors.

Balance low-latency ram with durable disk persistence by using the Chroma persistent client to upsert documents with embeddings and query with sqlite-backed storage.

Create OpenAI embeddings via the API using text embedding three small or large models, with a Python script and environment variables for the API key.

Switch to OpenAI embedding functions in Chroma database by configuring the API key and text embedding three small model, then index and query documents.

Assess chroma vector databases by monitoring latency, throughput, precision, recall, memory usage, and scalability, while leveraging inverted indexes, KD trees, HNSW graphs, and LSH to optimize real-time queries.

Build and explore vector databases through hands-on setup of VS Code, Python, and OpenAI API keys, then examine the chroma database workflow, the default embedding function, OpenAI embeddings, and metrics.

Discover how vector databases use similarity search in a vector space. Examine cosine similarity, dot product, and magnitude as measures of direction and semantic closeness between vectors.

Learn how Euclidean distance, the L2 norm, incorporates vector magnitude for clustering with k-means, and follow a concrete example showing the distance between (3,1) and (2,2) is about 1.41.

Explore the dot product as a key vector similarity measure used for image retrieval, music recommendation, and fraud detection, and see how it guides efficient database searches.

Identify the measures of vector similarity: cosine similarity for topic modeling and document similarity, Euclidean distance for clustering and anomaly detection, and dot product for image retrieval and music recommendation.

Load data into a vector store like chroma db, split documents into chunks, convert queries to embeddings, perform similarity search for relevant chunks, then use an llm to generate answers.

Load all articles from a data directory, split into chunks, generate OpenAI embeddings, and persist in a Chroma vector database to enable direct answers from a large language model.

Learn to split documents into chunks, generate OpenAI embeddings for each chunk, and upsert them into a chroma vector database for efficient search.

Comment out redundant setup, implement a query documents function for a vector database, generate embeddings, retrieve relevant chunks, then use a large language model to answer from those chunks.

Fetch relevant chunks, pass them with the user query to OpenAI's completion endpoint, and generate a concise answer using a context-driven system prompt and a vector store with embeddings.

Finish the workflow by embedding documents and splits, storing embeddings in the vector database, and retrieving similar entries to answer queries with a large language model like OpenAI.

Explore how vectors encode meaning as geometry, with 1536-dimensional embeddings that cluster documents and enable semantic search.

Discover why cosine similarity dominates text embeddings, versus dot product and Euclidean distance. Learn a practical framework to choose metrics for semantic search and normalized vectors.

discover how the LangChain framework enables plug-and-play, LLM-powered apps by uniting models, prompts, chains, retrieval, memory, and agents, with documents loaded and chunked in chroma.

Learn to set up LangChain, install it, and build a simple OpenAI chat application using the chat OpenAI wrapper with system and human messages to interact with GPT-4.

Learn to load documents with LangChain document loaders—directory and text loaders—and store them in a Chroma vector store using OpenAI embeddings and recursive text splitting.

Split the documents with LangChain using a recursive character text splitter to manage chunk size and overlap, yielding 223 split documents in small overlapping chunks.

Learn how to generate OpenAI embeddings with text-embedding-3, persist them into a chroma vector database via LangChain, and perform similarity searches through the stored documents.

Demonstrates querying a chroma vector store with a retriever to fetch relevant document chunks and metadata, then using LangChain to generate direct answers from a large language model.