Data and Search for AI Apps
Learn how AI apps store and find information: embeddings and vector search, hybrid ranking in PostgreSQL, time-series data, and securing agent tools.
Start learning Open course bookAbout this course
A language model on its own knows only what it was trained on. Useful AI applications surround it with data: documents to search, measurements arriving every second, and tools that let it act. How that data is stored, searched and protected decides whether the application gives accurate answers, stays fast as it grows, and keeps private information private.
This course works through that data layer in four steps. You begin with embeddings and vector databases, the technique behind search by meaning. You then combine keyword and vector search inside PostgreSQL and learn how the two rankings are merged. Next you handle time-series data, where rows arrive by the million and old data must be compressed and summarised. Finally you study the security of the Model Context Protocol, which connects AI agents to data and tools. Each lesson pairs a full video course with a book chapter built around figures, tables and worked calculations.
What you will learn
- Explain how text becomes an embedding and calculate cosine similarity by hand
- Describe what a vector index does and why its results are approximate
- Compare keyword ranking with vector ranking and merge them with Reciprocal Rank Fusion
- Explain how hypertables, compression and continuous aggregates keep time-series data fast
- Estimate row counts and storage for a stream of sensor readings
- Identify the main security risks when AI agents are connected to tools and data, and the defences against them
Who this course is for
Developers and technically minded learners who know basic Python and some SQL and want to understand the data side of AI applications. You do not need a machine learning background. If databases are new to you, expect to pause the videos often; the book chapters are written to stand on their own.