IBM's AI Engineering certificate now teaches LLMs, RAG and AI agents

IBM's AI Engineering Professional Certificate, already one of the most enrolled technical AI credentials on Coursera with over 259,000 learners, has expanded well beyond its original scope. What used to be a solid but fairly conventional deep learning course, covering neural networks, TensorFlow, and PyTorch, is now a 13-course series that spends roughly its second half entirely on generative AI.

This isn't a light refresh. It's a genuine restructure of what "AI engineer" is assumed to mean in 2026.

The two halves of the certificate

The first six courses are the traditional core: machine learning with Python, deep learning and neural networks with Keras, deep learning with Keras and TensorFlow, neural networks with PyTorch, and a capstone project applying CNNs and vision transformers to real image classification problems. If you took an earlier version of this certificate, this part will feel familiar.

The second half is new territory. Seven additional courses take learners through:

• Generative AI architectures and data preparation, covering RNNs, transformers, VAEs, GANs, and diffusion models, plus tokenisation for NLP with libraries like NLTK and spaCy.
• Foundational NLP and language modelling, including word embeddings, Word2Vec, and sequence-to-sequence models.
• Transformer-based language modelling, building the attention mechanisms behind models like GPT and BERT from the ground up in PyTorch.
• Fine-tuning, including parameter-efficient methods like LoRA and QLoRA, using Hugging Face and PyTorch.
• Advanced fine-tuning, covering instruction tuning, reward modelling, and reinforcement learning from human feedback (RLHF) techniques including direct preference optimisation and proximal policy optimisation.
• Building AI agents using retrieval-augmented generation (RAG) and LangChain, including tool calling and context engineering.
• A final project building a real generative AI application: a working question-answering bot with a vector database, a retriever, and a Gradio interface.

What this means in practice

The certificate now genuinely trains for two different jobs at once: the classic ML/deep learning engineer role, and the newer generative AI engineer role that involves fine-tuning and deploying LLM-based systems rather than training models from scratch. Coursera lists it as intermediate level, with four months of study at around ten hours a week as the expected pace, though the full 13-course sequence is substantial and self-paced learners can reasonably expect longer.

It's still squarely a technical, code-heavy program. IBM recommends working knowledge of Python and Jupyter Notebooks going in, and suggests completing the IBM Data Science or Applied AI certificates first if you don't already have that foundation.

Who this update is for

If you're a developer, data scientist, or ML engineer who already has the fundamentals and wants a structured path into building with LLMs specifically, fine-tuning, RAG, and agents rather than picking it up piecemeal from scattered tutorials, this expanded curriculum earns its place as one of the more comprehensive single certificates covering that ground. If you're after a gentle, non-technical introduction to generative AI, this was never the right course and still isn't, IBM's own AI Foundations content or Google's AI Essentials are a better starting point.

Bottom line

This is one of the more substantial curriculum expansions we've seen from an established certificate rather than a new course launched from scratch. The original deep learning core hasn't been diluted, generative AI engineering has been added as a genuine second half, which is a fair reflection of where technical AI hiring has actually moved.