IBM AI Engineering Professional Certificate

Learning plan

About 120 hours of focused study, split into 5 phases. Pick your weekly pace and the timeline adjusts.

Estimated finish: 12 weeks

What the exam covers

Source: official exam guide. Weights can change, so check before you start.

Before you start

• Working Python and pandas, plus high-school algebra and basic statistics
• A Coursera subscription. Labs run in IBM Skills Network browser environments, so no local GPU is needed

1. Weeks 1–2 · 15 h

   Classical ML with scikit-learn (course 1)

   • Train and evaluate regression, classification (logistic regression, KNN, SVM, decision trees) and clustering (k-means, DBSCAN) models with scikit-learn
   • Choose metrics for each task (RMSE, R², precision/recall, F1, silhouette score) and use cross-validation and regularisation

   Use: Machine Learning with Python · scikit-learn user guide

2. Weeks 2–6 · 45 h

   Deep learning with Keras and PyTorch (courses 2–5)

   • Build dense, convolutional and recurrent networks in Keras, including transfer learning with a pretrained CNN
   • Write PyTorch training loops by hand (Dataset, DataLoader, nn.Module, optimiser, loss), with dropout, batch normalisation and weight initialisation
   • Debug training with loss curves, learning-rate changes and GPU/CPU device handling

   Hands-on: In Google Colab, train the same small CNN on CIFAR-10 in both Keras and PyTorch. Match the architecture and compare accuracy, training time and lines of code.

   Use: Introduction to Neural Networks and PyTorch · Deep Learning with PyTorch · PyTorch tutorials

3. Weeks 7–8 · 12 h

   Deep learning capstone (course 6)

   • Load and augment a real image dataset, fine-tune a pretrained model and report validation accuracy
   • Compare a Keras and a PyTorch solution and justify the final choice in a written report

   Hands-on: Finish the capstone's image-classification project end to end and publish the notebook on GitHub with a README showing your metrics and confusion matrix.

   Use: AI Capstone Project with Deep Learning

4. Weeks 8–11 · 30 h

   Transformers, LLMs and fine-tuning (courses 7–11)

   • Explain tokenisation, embeddings, positional encoding, self-attention, and encoder vs decoder architectures (BERT vs GPT)
   • Build data-preparation pipelines with Hugging Face tokenizers and PyTorch DataLoaders
   • Fine-tune transformers with full fine-tuning, LoRA/QLoRA and adapters, and explain RLHF, PPO and DPO at a conceptual level

   Use: Generative AI Language Modeling with Transformers · Generative AI Advanced Fine-Tuning for LLMs · Hugging Face LLM Course (free)

5. Weeks 11–12 · 18 h

   RAG and agents with LangChain (courses 12–13)

   • Build a RAG pipeline with document loaders, text splitters, embeddings, a vector store and a retriever in LangChain
   • Add tools and a simple agent, and present the app through a Gradio interface
   • Evaluate retrieval quality and reduce hallucinated answers

   Hands-on: Build a "chat with your PDFs" app using LangChain, a local vector store and Gradio. Test it with 10 questions whose answers you know, and log which ones it gets wrong and why (retrieval miss or generation error).

   Use: Fundamentals of AI Agents Using RAG and LangChain · Project: Generative AI Applications with RAG and LangChain · LangChain RAG tutorial

You're ready to book when…

• All 13 courses are complete, with every graded project submitted and passed
• You can explain self-attention and LoRA on a whiteboard
• Your capstone and RAG app are public on GitHub with clear READMEs

Exam tips

• Coursera's listed course times add up to about 168 hours, more than the program's headline estimate. If you already know scikit-learn, skim course 1 and keep your hours for the PyTorch and GenAI courses.