Hands-On Training: AI Workflows with Postgres, Vectors, and JSON

Who should attend

This course is aimed at Data Scientists, Analysts, Database Developers, Data Warehouse Managers, Business Intelligence Specialists, Software Developers and Architects.

Prerequisites

You should be comfortable with the basics of SQL, ideally with experience of using Postgres database. We can help you get up to speed with a primer session as part of a custom workshop.

Live, instructor-led online and on-site training

This Postgres Vector Store training course is available for private / custom delivery for your team - face-to-face, on-site at your location of choice, or remotely via MS Teams or your own platform - get in touch to find out how we can deliver tailored training which focuses on your project requirements and learning goals.

Overview of AI Workflows

• Look at high-level architecture - LLMs, vector stores and JSON
• Look at key vocabulary and concepts (embeddings, )
• Why Postgres as a vector store

What is a vector store? Key concepts and use cases

• Why Postgres and how does it compare with other market tools
• Setting up Postgres with vector capabilities (pgvector)
• Lab: Install and configure Postgres using Docker

Storing and managing vectors

• Generating embeddings: Overview of tools and workflows (e.g. OpenAI, Hugging Face)
• Storing and organizing embeddings in Postgres
• Lab: Generate embeddings for a dataset and store them

Querying the vector store

• Techniques for similarity search: k-NN, cosine similarity
• Using indexes to optimize vector queries
• Lab: Query stored vectors to retrieve similar items (document/image search)

Querying LLMs with retrieved data

• Recap on querying LLMs vis APIs
• Best practices for combining vector retrieval with LLM prompts
• Lab: Build a pipeline where vector store results enhance LLM responses (context-aware Q&A, etc)

NoSQL with JSON in Postgres

• Overview of JSON/JSONB support in Postgres
• Querying JSON data with SQ:
• Indexing JSON data for performance
• Lab: Design a schema mixing vector, relational and JSON data for a sample project

Integrating Vector, Relational and JSON Data

• Building hybrid queries to power advanced workflows
• Case study: Combining embeddings, metadata (relational) and configurations (JSON)
• Lab: Implement a hybrid query to support a sample AI use case

Putting it all together

• Full stack pipeline demo: Retrieve data, query the LLM and return results
• Debugging and optimising the workflow
• Lab: Build a working application combining all elements