← All courses

🤖 AI for Data

Text-to-SQL you can trust, embeddings and vector search, RAG grounded in your own schema — AI as a data copilot, on your data.

No lessons published yet — this is the roadmap, so you can see exactly what the course will cover before it lands.

Before you start: Learn SQL — Series 2. Every episode in Series 1 ends with you checking a generated query. A model that invents a column or writes the wrong join produces a plausible number, and multi-table joins are exactly where it goes wrong — so Series 2, not Series 1.

Short on time? The 20 episodes marked core are the shortest path to being productive at work. Start there and pick up the rest whenever a job asks for them.

Series 1 · AI as Your Data Copilot

use AI to write SQL, and check what it gives you

  1. 01

    What an LLM Can and Cannot Do With Your Data — Where It Helps and Where It Guesses

    corebeginnercoming soon
  2. 02

    Text-to-SQL — Ask in English, Then Read the Query It Wrote

    corebeginnercoming soon
  3. 03

    Verify the Answer — Checking the Generated Query Returns the Right Rows

    corebeginnercoming soon
  4. 04

    Schema Context — Why the Model Needs Your Tables, Not Just Your Question

    corebeginnercoming soon
  5. 05

    Prompt Patterns for Data — Few-Shot, Constraints and Guardrails

    corebeginnercoming soon
  6. 06

    Failure Modes — Invented Columns, Silent Wrong Joins, Confident Nonsense

    corebeginnercoming soon
  7. 07

    AI in the Query Editor — Explain, Refactor and Optimize an Existing Query

    beginnercoming soon
  8. 08

    Privacy & Local Models — Keeping Your Data Off Someone Else's Server

    corebeginnercoming soon
  9. 09

    Documenting With AI — Schemas, Lineage and Runbooks You'll Actually Keep

    beginnercoming soon
  10. 10

    Project — Answer a Real Northwind Question With AI, and Prove It's Right

    corebeginner#projectcoming soon

Series 2 · Embeddings & Retrieval

embeddings, vector search and RAG on your own data

  1. 01

    Embeddings — Turning Rows and Text into Vectors

    coreintermediatecoming soon
  2. 02

    Similarity & Distance — Cosine, Euclidean and Approximate Nearest Neighbours

    intermediatecoming soon
  3. 03

    Vector Storage Where Your Data Already Lives — pgvector and DuckDB vss

    coreintermediatecoming soon
  4. 04

    Semantic Search Over Northwind — Find Products by Meaning, Not LIKE

    coreintermediatecoming soon
  5. 05

    Chunking & Metadata — The Settings That Decide Whether Retrieval Works

    intermediatecoming soon
  6. 06

    RAG — Retrieve First, Then Generate

    coreintermediate#lewis-ragcoming soon
  7. 07

    RAG on Your Warehouse — Ground the Model in Your Schema and Docs

    coreintermediatecoming soon
  8. 08

    Hybrid Search — Combining Keyword and Vector Retrieval

    intermediatecoming soon
  9. 09

    Evaluating Retrieval — Recall, Precision and a Test Set You Trust

    coreintermediatecoming soon
  10. 10

    Project — A Question-Answering Layer Over Your Own Data

    coreintermediate#projectcoming soon

Series 3 · Agents, Evaluation & Production

give an AI tools, limit what it can touch, and test it

  1. 01

    Tools & Function Calling — Let the Model Run a Query, Under Supervision

    coreadvancedcoming soon
  2. 02

    Agents Over the Warehouse — Multi-Step Questions, Bounded Permissions

    advancedcoming soon
  3. 03

    Read-Only Contracts & Least Privilege — What an Agent Must Never Touch

    coreadvancedcoming soon
  4. 04

    Eval Harnesses — Score Generated SQL Against Known-Good Answers

    coreadvancedcoming soon
  5. 05

    Regression-Testing Prompts — Keep Quality When the Model Changes Under You

    advancedcoming soon
  6. 06

    Cost & Latency — Context Size, Caching and What to Precompute

    advancedcoming soon
  7. 07

    Fine-Tuning vs RAG vs Prompting — The Cheapest Thing That Actually Works

    advancedcoming soon
  8. 08

    Structured Output — JSON, Schemas and Parsing You Can Rely On

    coreadvancedcoming soon
  9. 09

    Observability for AI Features — Log the Prompt, the Query and the Answer

    advancedcoming soon
  10. 10

    Project — Ship an AI Data Feature: Grounded, Evaluated, Auditable

    coreadvanced#projectcoming soon

Not here, on purpose

  • Classical machine learning — features, training, model selection — not covered; this track is LLM-era work on data you already have
  • Reading and fixing the SQL a model writes — covered in Learn SQL — you still have to be able to check its work
  • Dashboards and metric governance — covered in BI & Data Visualization
  • Training or fine-tuning a model of your own — not covered; Series 3 · 30 tells you why RAG is almost always the cheaper answer, and stops there
  • Chatbots, agents and prompt engineering in general — not covered as a subject in itself — everything here is pointed at data you are responsible for
  • Running the vector store — sizing, backups, index rebuilds — covered in Data Ops; here pgvector is switched on inside the PostgreSQL DataPallas already ships