Course 3 For members Pilot
Building a RAG system
For AVI members. Open to everyone during the pilot.
- Who it is for
- Informatics-minded members and career changers
- Length
- About 11 hours
A hands-on course where you build a working retrieval-augmented generation (RAG) system over veterinary text, then learn to judge whether it gives good answers. It takes about eleven hours, in hosted notebooks, so there is nothing to install.
Who it is for¶
Veterinary professionals and informatics people who are comfortable with a computer but are not software engineers. You can follow instructions in a notebook and are willing to read a little Python. You do not need a machine learning background.
What you leave with¶
A working RAG project with test results, and the words to explain it in a job interview.
Start here¶
The opening module, How LLMs work, is open to everyone. It is shared with Using AI in practice.
Optional background Free
How AI works
The free tour explains the background to this course in more depth, in five short modules of interactive diagrams. Take it before the course or alongside it. It is not a step in the course.
- Module 1 Neural networks About 20 minutes
- Module 2 Language models About 20 minutes
- Module 3 Retrieval About 15 minutes
- Module 4 Agents About 15 minutes
- Module 5 Images About 15 minutes
The modules¶
Modules 0 to 4 are open now, as a pilot draft. Each module is a notebook that runs in Google Colab, in your browser, with nothing to install. Open a notebook straight away, or read its page first: each page says what you will build and how long it takes.
-
The opening module
How LLMs work
You don't need any math to use AI tools well, but you do need a working picture of what they are doing. This module gives you that picture, and shows why the same tool can be useful and wrong in the same answer.
-
Module 0
Setup and notebook primer
Your starting line. You open a Google Colab notebook in your browser and run your first lines of Python: variables, text, lists, and how to read an error. Then you get a free Gemini key, store it safely in Colab's Secrets, and get your first answer from a model. You need no Python experience, and you install nothing on your computer.
-
Module 1
Why RAG exists
See the problem before you build the fix. You ask a model ten questions about a made-up clinic's protocols, with no documents to help it: a drug dose, a protocol that changed, a species trap, clinic shorthand, and one question nothing covers. You record where it goes wrong on a worksheet you keep, and meet the three ways to give a model your own documents.
-
Module 2
Documents, chunking and embeddings
Turn documents into something a computer can search by meaning. You pull the text out of clinic protocols and a real journal PDF, see what happens to its tables, and write a function that cuts text into overlapping chunks. Then a small open model, running inside the notebook, turns each chunk into an embedding, and you test whether search understands clinic shorthand such as parvo and HBC.
-
Module 3
Retrieval
Build the search that feeds the assistant. You compare keyword search (BM25), search by meaning and a hybrid of the two, see why drug names favour keywords, and fix clinic shorthand with an abbreviation table. Then you look at what retrieval returns for all ten questions before any model sees them: the habit that matters most in this course.
-
Module 4
Generation with citations
Put it together into an assistant you can check. You write the prompt that makes a model answer only from the retrieved passages, cite each one, and say plainly when the documents do not cover a question. Then you rerun the ten questions from Module 1 and compare: what retrieval fixed, and what it did not.
-
Module 5 Coming soon
Evaluation
Build a test set and score retrieval and answers.
-
Module 6 Coming soon
Failure modes and safety
Break the assistant on purpose, then fix it.
-
Module 7 Coming soon
Capstone and teach back
Build over your own documents and show it to others.