Code course: AI apps

AI engineering course: build, test and run apps on AI model APIs

This AI engineering course teaches you to build small applications on top of AI model APIs: sending requests, getting reliable structured output, searching your own documents by meaning, answering from them with citations and proving the results are good enough to use. It works as an LLM course and a RAG course for people who already know basic Python and want to build AI apps responsibly.

For you or your whole team

Start with a useful first step.

7-day free trial. No credit card needed.

You will continue on the Tosheo.ai app to create your account and review the trial terms.

What you will learn

The modules in this course.

Calling model APIs

Requests, responses and API keys kept safely outside your code.

Model choice and cost

Picking a model per task and estimating cost per use.

Structured output

Reliable JSON, validated in code before it is used.

Embeddings and RAG

Searching documents by meaning and answering with citations.

Evaluation

Measurable success criteria and test sets that catch regressions.

Safety and monitoring

Prompt injection defences, private logs, rate limits and quality tracking.

Example task / fictional source material

Capstone: a document question-answering app with an evaluation report

Build a small app that answers questions from a set of public or made-up documents in your field, such as published regulations or invented product manuals. Include citations, an evaluation set of at least fifteen questions and a one-page report on quality, cost and known limits.

Review criterion

Grounded answers

Answers cite the passages used, and the app declines when the documents do not contain the answer.

Review criterion

Real evaluation

The test set includes easy, hard and unanswerable questions, with scores recorded before and after one improvement.

Review criterion

Safe handling

API keys are outside the code, logs hold no personal data, and one prompt injection attempt is tested.

Review criterion

Honest report

The report states cost per question, failure cases and what a reviewer should check before wider use.

Copy and adapt

Prompts to try

Replace the parts in square brackets with your own details, and check every output before you use it.

Design a JSON schema

I am building [feature]. Propose a JSON schema for the model's output with field names, types and allowed values, and list three invalid responses my validation code should reject.

Write evaluation cases

For a helper that answers questions from [document type], write twelve test questions: six answerable, three hard and three that the documents cannot answer. Give the expected behaviour for each.

Find injection risks

Here is how my app builds its prompt: [description without keys or real data]. List the places where untrusted text could change the model's instructions and suggest one defence for each.

Every lesson

14 lessons you can start today.

Your diagnostic picks the order and skips what you already know. Start any lesson with a 7-day free trial, no credit card needed.

How AI APIs workSend a request to a model API and read the response
PractitionerCodeStart this lesson
Choosing a model and managing costCompare models on quality, speed and price for a task
PractitionerCodeStart this lesson
Structured outputGet reliable JSON from a model
PractitionerCodeRead the free lesson Start this lesson
Embeddings and semantic searchFind similar documents by meaning
PractitionerCodeStart this lesson
Retrieval-augmented generation (RAG)Answer questions from your own documents with citations
AdvancedCodeRead the free lesson Start this lesson
Context engineering in appsDecide what goes into each model call
AdvancedCodeStart this lesson
Prompt and response evaluationBuild a test set and score outputs
AdvancedCodeStart this lesson
Streaming and user experienceShow answers as they generate
PractitionerCodeStart this lesson
Image and audio APIsAdd transcription, speech or images to an app
PractitionerCodeStart this lesson
Building a simple AI web appShip a small app with a front end and a model call
AdvancedCodeStart this lesson
Safety, moderation and abuseAdd input checks, output filters and rate limits
AdvancedCodeStart this lesson
Privacy and data handlingKeep user data out of logs and training
AdvancedCodeStart this lesson
Monitoring an AI featureTrack errors, costs and quality after launch
AdvancedCodeStart this lesson
Answer engine optimisation (AEO)Make content easy for AI assistants to find and cite
PractitionerCodeStart this lesson

Who is this AI engineering course for, and who is it not for?

It is for analysts, technical product managers, consultants, data people and junior developers who can write basic Python and want to add AI features to internal tools: a document question-answering helper, a classifier for support emails, or a summariser that returns data a spreadsheet can use.

It is not a full-time bootcamp, a machine learning degree or a job guarantee. You do not train or fine-tune models, and you do not study the mathematics of neural networks. The focus is the engineering around a model: inputs, outputs, cost, quality, safety and privacy.

What will I learn when I build AI apps on this course?

Six modules, each producing a working piece of a larger app. Descriptions of provider features follow the providers' own developer documentation, and features differ between providers.

1. How AI APIs work

Sending a request to a model API from Python, reading the response and keeping the API key in an environment variable, not in your code.

2. Choosing a model and managing cost

Comparing models on quality, speed and price for one task, using each provider's own pricing page, and estimating what a feature will cost per use.

3. Structured output

Getting reliable JSON. OpenAI documents Structured Outputs as making responses adhere to a JSON Schema you supply; you still validate every response in code.

4. Embeddings and retrieval-augmented generation

Embeddings measure how related pieces of text are. You use them to find relevant passages, then build RAG: answering from your own documents with citations.

5. Evaluation

Writing specific, measurable success criteria and a test set that mirrors real tasks, including edge cases, then scoring outputs automatically where possible.

6. Safety, privacy and monitoring

Defending against prompt injection, keeping user data out of logs, adding input checks and rate limits, and tracking errors, cost and quality after release.

Do I need experience before this LLM course?

Yes. The prerequisite is the Tosheo.ai Python course or equivalent: you should be able to write functions, use lists and dictionaries, read and write files, install a package in a virtual environment and call a web API. If the diagnostic shows gaps, it adds the missing Python lessons to the start of your path rather than letting you struggle. You do not need any machine learning background.

How does the personalised path adapt to me?

The diagnostic asks you to read a short script that calls a model API and to say what goes wrong when the response is not valid JSON. It also asks what you want to build. Someone aiming at document search spends longer on embeddings and RAG; someone building a classifier goes deeper into structured output and evaluation.

If you can already make API calls, module one shrinks to a quick check, and you begin at cost and model choice.

What do I practise?

Each task is code you run against a real model API with made-up data. One example from module four: Kestrel Lettings, a fictional property management firm, has fifteen invented policy documents for landlords. You build a small RAG helper that answers questions such as how much notice is needed for an inspection, quoting the passage it used.

The rubric is visible before you begin. It checks that every answer cites a passage that really contains the answer, that the helper says it does not know when the documents are silent, that you tested at least ten questions including three with no answer in the documents, and that you recorded the cost of the test run.

How long does the AI engineering course take?

Lessons last 10 to 15 minutes each, with code kept short enough to finish in one sitting. The number of lessons in your path depends on your diagnostic, especially your Python, and on your pace, so we do not give an hours figure.

What do I get at the end?

The final practical assessment is a small AI feature you build, evaluate and document. Passing earns a recognised certificate. It is not an accredited qualification and is not issued or endorsed by any model provider.

How much does the AI engineering course cost?

Tosheo.ai's Individual plan, which includes this course, is USD 79 monthly or USD 790 yearly. The Team plan is USD 189 monthly or USD 1,890 yearly for up to 3 members; extra members cost USD 49 monthly or USD 490 yearly, and teams of 25 or more get volume pricing. Model providers charge separately for API usage, and the course shows you how to keep test runs small. Try it first with a 7-day free trial, no credit card needed.

Sources and scope

OpenAI API docs, Structured Outputs, checked 19 October 2026 ↗

OpenAI API docs, Embeddings, checked 19 October 2026 ↗

OpenAI API docs, Data controls in the OpenAI platform, checked 19 October 2026 ↗

Claude Docs, Define success criteria and build evaluations, checked 19 October 2026 ↗

Claude Docs, Reduce hallucinations, checked 19 October 2026 ↗

OWASP Top 10 for LLM Applications 2025, LLM01 Prompt Injection, checked 19 October 2026 ↗

Before you decide

A few good questions.

What is AI engineering?

AI engineering is building software around AI models: choosing a model, shaping its inputs and outputs, grounding it in data, testing quality and running it safely. It differs from training models.

Is this a RAG course?

Yes, in part. Retrieval-augmented generation has its own module, with embeddings, citations and testing, and it is also the subject of the capstone.

Which AI provider does the course use?

Lessons explain concepts that apply across providers and show examples using official APIs. Provider features differ, so each lesson links to the provider's own documentation.

Will my test data be used to train models?

It depends on the provider and its terms. OpenAI, for example, says API data is not used to train its models unless you opt in. The course teaches you to check this for every provider you use.

Can I take this LLM course without the Python course?

Yes, if you already have equivalent Python skills. The diagnostic checks them and adds any Python lessons you need first.

Give your team a path, not another playlist.

Explore role-specific lessons, practical tasks and the manager view.

Start free trial7-day free trial · No credit card needed

AI engineering course: build, test and run apps on AI model APIs

Start with a useful first step.

The modules in this course.

Calling model APIs

Model choice and cost

Structured output

Embeddings and RAG

Evaluation

Safety and monitoring

Capstone: a document question-answering app with an evaluation report

Grounded answers

Real evaluation

Safe handling

Honest report

Prompts to try

Design a JSON schema

Write evaluation cases

Find injection risks

14 lessons you can start today.

Who is this AI engineering course for, and who is it not for?

What will I learn when I build AI apps on this course?

1. How AI APIs work

2. Choosing a model and managing cost

3. Structured output

4. Embeddings and retrieval-augmented generation

5. Evaluation

6. Safety, privacy and monitoring

Do I need experience before this LLM course?

How does the personalised path adapt to me?

What do I practise?

How long does the AI engineering course take?

What do I get at the end?

How much does the AI engineering course cost?

Sources and scope

A few good questions.

Keep exploring.

Python course to meet the prerequisite

Build AI agents in code and no-code

Machine learning course for professionals

Browse all Tosheo.ai courses

Tosheo.ai plans and prices

Give your team a path, not another playlist.