Country or region and language

    Choose your country or region and your language.

    Suggested for you

      All countries and regions

        Choose your character

        It shows in the top bar and on your profile. Change it any time.

        Courses / AI tools and development

        AI engineering course with Python, RAG and agents

        Write the Python around a language model: validate an API response, retrieve the right source, run an approved tool and check the result. This AI engineering course runs your code against scripted mock replies in a browser notebook, with no live model calls.

        Start lesson 1Practice interviewTake the assessmentPractice the first unit of every course. No account needed.
        60lessons
        338checked exercises
        12projects, cases and exams
        Prepares you for Engineering interviews and original timed job assessments

        What you will be able to do

        • Validate structured output, refusals and incomplete responses
        • Build a local retrieval pipeline with tenant-scoped sources and citations
        • Execute allowed tools with checked arguments and bounded loops
        • Compare frozen evaluation cases and preserve safe operational evidence

        Who it is for

        You can write Python functions
        Build and test the application code that surrounds a language model.
        You review a model-backed feature
        Check source boundaries, unsafe actions and evaluation evidence before a release.

        What you can practice free

        Unit 1, Request contracts: 5 lessons and 26 checked exercises, with no account or card.

        Start the free unit

        The rest of the course opens with a One Course or All Access plan. See plans and prices

        Keep going in Building with language models

        Choose access here. Your course is already selected.

        One Course

        Every lesson, project and mock exam in Building with language models.

        Loading price/ month

        Billed monthly. Cancel any time.

        All Access

        Every course, exam and project, plus the Interview Hub.

        Loading price/ month

        Billed monthly. Cancel any time.

        Projects, case studies and mock exams

        Build language-model features in a working Python notebook with an original local mock client. Construct current Responses-style requests, validate structured output and correlate tool results. Build traceable retrieval, compare prompt revisions, bound agent loops and preserve tenant and approval boundaries. Projects accumulate actual learner functions into checked reports. Practice sets, interviews and two timed job screens use original fictional data and synthetic observations. The mock performs no live inference or external tool execution. Its documented API subset, fake vectors and fake token units teach integration behavior without claiming model quality, provider prices or complete cloud certification preparation.

        Beginner

        FreeNo account or card needed. Every lesson in this unit is yours to practice.

        Request contracts

        Support routing requests with explicit authority and resource boundaries.

        5 lessons

        Structured output

        Validate output shape, response state and business rules before accepting an answer.

        5 lessons

        Tool round trips

        Define narrow tools and return correlated results without granting unrequested actions.

        5 lessons

        Intermediate

        Retrieval index

        Build an original industrial-manual index with traceable chunks and deterministic vector fixtures.

        5 lessons

        Grounded answers

        Answer customer policy questions only from current authorized evidence.

        5 lessons

        Evaluation harness

        Compare prompt revisions with frozen case identities and mandatory safety gates.

        5 lessons

        Advanced

        Agent loops

        Prepare a bounded dispatch workflow with exact-payload approval and reconstructable state.

        5 lessons

        Memory and safety

        Investigate a tenant leak while preserving safe incident evidence.

        5 lessons

        Resilient integration

        Recover transient and uncertain outcomes without repeating unauthorized work.

        5 lessons

        Expert

        Cost and speed

        Build a batch report with safe cache identity, output correlation and supplied observation gates.

        5 lessons

        Operational evidence

        Trace a grounded feature and prepare a reproducible release bundle.

        5 lessons

        Production decisions

        Enforce approved data policy and decide release or rollback from preserved evidence.

        5 lessons

        Request and output contracts

        A support feature should not parse the first returned item as if it were always text. Inspect the Responses API output items, distinguish a tool request from a message and handle an incomplete result. The structured output exercises add a schema, then require business validation. Valid JSON can still be wrong.

        Retrieval and evidence

        Build a RAG pipeline for fictional policy documents. Filter before ranking. Use only the requesting customer’s sources at the current revision. Check that every citation belongs to the retrieved evidence and abstain when the sources do not support an answer. The vector search guide explains what the local fixture simplifies.

        Use held-out evaluations to compare a candidate with a fixed baseline. Keep the cases used to improve the code separate from those used to report its result. A higher average does not excuse a failed mandatory safety case.

        Tools and agent loops

        Correlate function results by call ID, validate arguments and reject unknown tools. Add a step budget and require approval for the exact write payload. The prompt injection exercises include hostile source text; the application must enforce access even when the assistant reply is wrong.

        Compare RAG and fine-tuning and MCP and function calling before choosing a live architecture. These concept guides do not install a protocol server or train a model.

        Interviews and credentials

        Practice original engineering questions in the Interview Hub. The AI-103 guide explains the current Azure credential and the retirement of AI-102. The AWS AI Practitioner guide separates foundational certification knowledge from coding a model application. Our timed job assessments do not reproduce either exam.

        Common questions

        Does this course call a live model?

        No. Your Python runs against an original mock client with prepared outputs. No API key or paid model account is needed, and a passing task does not measure a live model.

        Does the retrieval exercise use a real embedding service?

        No. You build deterministic fixture vectors and calculate similarity locally. They teach indexing and ranking behavior; they do not reproduce the quality of a learned embedding model or a managed vector database.

        Is this an AI-103 or AWS certification course?

        It is practical engineering training. The certification guides map selected concepts and explain the gaps. The timed assessments are original job tasks, not vendor exam simulations.

        Do I need Python first?

        You should be able to write functions, use dictionaries and lists, and handle exceptions. Start with the Python course if those tasks are new.

        Your Work Gym account

        Save your progress and pick up on another device. Your browser's progress comes with you.

        or

        You can keep practicing without an account.

        Change your plan

        Checking the price

        Help and feedback

        Ask a question, tell us what is not working, or share an idea. We reply by email.

        What is it about?

        We use it only to reply to you.