Introduction to AI Engineering
Why it matters if you work with Python: The map of the field, from LLMs and embeddings to agents and evaluation, with its code examples mostly in Python, so you know which part of the stack you want to specialise in.
Python is the language of PyTorch and of the agent frameworks and evaluation tools these courses teach, and both OpenAI and Anthropic publish an official SDK for it. That gives you a head start, not a result: the difference between a Python developer who has "tried the API" and one who ships AI systems in production shows in architecture, evaluation and operations, not in syntax.
In the order in which they matter, each with the reason it matters if you work with Python. Untick what you already know: the subscription and the price update.
You start with what you already know, Python, and add the mental model of AI systems: how an LLM works, what it can and cannot do, and how you call it properly from code.
Why it matters if you work with Python: The map of the field, from LLMs and embeddings to agents and evaluation, with its code examples mostly in Python, so you know which part of the stack you want to specialise in.
Why it matters if you work with Python: The modules on prompt engineering for code and for agents show how to write instructions that produce verifiable code and decisions, not just pleasant text.
Why it matters if you work with Python: Streaming, error handling, function calling, multi-model routing and productionization, with the examples mostly in Python: the part notebooks skip and production demands.
This is where the difference is made: you turn API calls into systems with your own data, memory, tools and predictable behaviour.
Why it matters if you work with Python: Chunking, embeddings, hybrid search, re-ranking and the systematic evaluation of a RAG pipeline, built with the Python frameworks the field uses: LangChain, LlamaIndex and Haystack.
Why it matters if you work with Python: Agents with memory, state and tools, design patterns for production, guardrails and observability, built with frameworks and SDKs from the Python ecosystem such as LangGraph, the OpenAI Agents SDK and CrewAI.
Why it matters if you work with Python: A whole module on building MCP servers with Python, using FastMCP: your tools become available to any compatible client.
Why it matters if you work with Python: Memory types, retrieval for context and compaction, with the examples in Python: what you do when the context window is no longer enough and the agent "forgets".
An AI system ships with metrics and has to be operated. This is the stage where you answer for quality, for the models you train yourself and for how they run.
Why it matters if you work with Python: promptfoo, DeepEval, Ragas and pytest in practice: golden datasets, a calibrated LLM-as-a-judge and regression tests in CI/CD, so you stop deploying quality regressions.
Why it matters if you work with Python: Built entirely on PyTorch: tensors, autograd, training loops, CNNs and Transformers, GPU training and ONNX export. The layer under the APIs, in your language.
Why it matters if you work with Python: For when prompting and RAG are not enough: datasets, LoRA and QLoRA with the Hugging Face toolchain, evaluation, quantization and self-hosting with Ollama and vLLM.
Why it matters if you work with Python: Experiment tracking, training pipelines, a model registry, serving and drift monitoring: the lifecycle around the models you train, for when your project has trained models and not only API calls.
Why it matters if you work with Python: Pipelines, data quality, embeddings at scale and vector databases, with Python and SQL throughout: the data layer every retrieval system depends on.
Why it matters if you work with Python: Agentic coding from the terminal on real codebases: plan, edit, review, deterministic hooks, CI and least-privilege permissions, for the Python projects you already maintain.
Not sure what you can skip? Take the free level assessment
The course in the track that ends with a capstone project, from their real modules.
It opens next to the lesson and starts from it. You type or talk to it, and you can listen to the answer.
Included in every plan and in single-course access, within a fair-use limit. It is an artificial intelligence system, not a human teacher; its answers may contain errors. Details in the Terms.
The professor opens next to the lesson, on the section you have reached, with four ways to start: a summary, applying it to your job, mistakes to avoid, or a test.
You select a passage and ask for an explanation. The answer starts from the lesson, and you can listen to it, copy it or save it to your notes.
The ideas of the lesson, point by point, when you ask for them. You can listen to the summary or send it to your notes.
3, 5 or 10 new questions from the lesson you are reading, with an explanation for every answer.
Flashcards made from the lesson: you flip the card and mark what you already knew.
Screenshots from the app, taken in the demo lesson, with prepared answers.
The position of the language, what changes in your work, the tools and what you will be able to do at the end.
Much of what is new in applied AI is available in Python: OpenAI and Anthropic publish official Python SDKs, the Model Context Protocol has an official Python SDK, and PyTorch, the Hugging Face libraries, inference servers such as vLLM and evaluation tools such as DeepEval and Ragas all live here. If you write Python, you do not have a foreign ecosystem to learn. You have to learn to use this one properly.
The risk for a Python developer is different from the one your Java or .NET colleagues face: not missing tools, but shallowness. A notebook with one API call and a long prompt is not an application. What sets you apart is being able to build a RAG pipeline you can evaluate, an agent with memory and guardrails, and a service with retries, cost under control and regression tests on answer quality.
A language model is a non-deterministic component with high latency and variable cost. Streaming, tool use, structured outputs, retries and token budgets are now part of everyday work, the way database connections used to be.
Embeddings, chunking, vector databases and retrieval strategies enter the usual stack. A Python developer who understands RAG can connect an internal source to a model and keep control over what the model "knows".
It is no longer enough for the function to return without an exception. You need golden datasets, metrics for retrieval and generation, a calibrated LLM-as-a-judge and regression tests in CI; otherwise every prompt change is an invisible regression.
Cursor, Claude Code and terminal agents work on Python codebases. Whoever knows how to give them context, limit their permissions and verify their output ships faster without losing control of the code.
These are the skills you practise in the courses of the track; results depend on the time you invest and on your previous experience.
Start the Python trackThe whole Python track fits in a subscription of 15 courses: €359 a month, VAT included. You pick the courses and can change them at every renewal.
Pay for the tier, then pick the courses in your account. You can change them at every renewal.
€99 a course
€99a month €990a year, 12 months of access
under €67 a course
€199a month €1,990a year, 12 months of access
under €52 a course
€259a month €2,590a year, 12 months of access
under €32 a course
€319a month €3,190a year, 12 months of access
under €24 a course
€359a month €3,590a year, 12 months of access
under €16 a course
€399a month €3,990a year, 12 months of access
under €10 a course, across all 50
€499a month €4,990a year, 12 months of access
Total prices, VAT included. Move up anytime and pay only the difference for the days left; move down at renewal. Renews automatically, you stop it anytime from your account; your progress and certificates stay.
Payments are processed by Stripe, a PCI DSS Level 1 certified provider, with 3-D Secure for payer authentication. The connection to this site is TLS-encrypted.
Compare all the subscriptions · The team library, priced per member · Give a month of access as a gift
When you finish the lessons of a course with an average of at least 70% on its quizzes, you issue the PDF certificate from your account, at no extra cost.
A private attestation of course completion. It is NOT a diploma or a state-recognised qualification under Romanian Laws 198/2023, 199/2023 or Government Ordinance 129/2000; employers may take it into account at their own discretion.
The same skills, seen from each ecosystem. The languages close to Python are highlighted.
Most of them use Python for their code examples: the introduction, LLM integration, RAG, agents, context engineering, evaluation, PyTorch, fine-tuning and data engineering. The MCP course has one module on servers in Python and one on servers in TypeScript. A few courses, such as MLOps, AI security and the AI SaaS course, explain architecture and practice in prose rather than through code listings.
Not for most applications. Integrating APIs, RAG, agents and evaluation are software engineering, not research. The introduction to AI engineering has a module on the machine learning essentials you need to make informed decisions. PyTorch, fine-tuning and MLOps are optional steps, for when the project needs its own models.
Skip the foundations stage and start with RAG or with advanced LLM integration, then continue with agents and evaluation. Every course has a free preview of its first lesson, readable without an account, so you can check the level before you subscribe.
All the courses on this page are IT courses. You pick them in a subscription by number of courses: usually 3, 5 or 10 courses a month for one language and stack; there are also 15 and 25 courses, a single course and the full catalog. You pick the courses and can change them at every renewal. The subscription renews automatically; you can stop the renewal anytime from your account, and access stays until the end of the period already paid for. The prices, with VAT included, are shown below and on the pricing page.
13 courses, ~327 hours, in a subscription of 15 courses: €359 per month, VAT included. You can change them at every renewal.
The courses are for educational purposes only: they are not professional advice, and completing them is not a diploma or a state-recognised qualification. The names of languages, tools and products belong to their owners and are mentioned for identification only. Details in the Terms & Conditions.
ChatGPT, OpenAI, Claude, Anthropic, Gemini, Google, Microsoft 365 Copilot, Cursor, Midjourney and the other product and company names mentioned are trademarks or trade names of their respective owners. Cursuri AI is an independent training provider and is not sponsored or endorsed by any of them; our membership in Anthropic's Claude for Startups program does not mean that Anthropic endorses our courses.