Generative AI and MLOps. Build AI systems, then run them in production.

How machines read and represent text.

Topics covered

• Tokenisation and text preprocessing
• Embeddings: dense representations of words and images
• Text classification with deep learning

You produce: a text classification model.

How machines process and recognise images.

Topics covered

• Image processing: resizing, filtering and normalisation
• Image classification with deep learning
• Facial recognition: detection, recognition and verification

You produce: an image classification model.

The architecture behind modern language models.

Topics covered

• Transformer architecture and attention
• BERT: bidirectional encoder representations
• GPT: generative pre-trained, autoregressive language models

You produce: a working comparison of BERT and GPT.

Adapting language models to real tasks.

Topics covered

• Named entity recognition (NER)
• Sentiment analysis
• Fine-tuning pre-trained models
• Prompt engineering: crafting effective prompts

You produce: a fine-tuned model for an NLP task.

Generating text, images and multimodal output.

Topics covered

• OpenAI APIs: text generation, embeddings and DALL·E
• Gemini APIs: multimodal interaction and reasoning
• Multimodal models: text, image and audio
• Generative adversarial networks (GANs)

You produce: a multimodal generative AI prototype.

Grounding LLMs in your data and building agents.

Topics covered

• Retrieval-augmented generation (RAG) pipelines
• LangChain: chaining LLM calls, tools and memory
• LangGraph: graph-based state management for agents
• A2A protocol and MCP (Model Context Protocol) for multi-agent systems

You produce: a RAG-powered question-answering system.

Tracking, versioning and packaging models.

Topics covered

• The ML lifecycle: data, training, deployment and monitoring
• Model tracking and packaging with MLflow
• Versioning data, code and models with Git and DVC
• Docker for ML applications

You produce: a versioned, tracked ML experiment.

Automating training, testing and deployment.

Topics covered

• Pipeline automation and orchestration: Airflow, Kubeflow Pipelines, Prefect and Argo
• Kubeflow for production ML workflows on Kubernetes
• CI/CD for ML: automated testing and deployment
• Data drift detection and performance alerts

You produce: an automated MLOps pipeline.

Putting a model behind an API or an interface.

Topics covered

• Flask for lightweight model services
• FastAPI for high-performance async APIs
• Streamlit dashboards and demos
• Scalable APIs: authentication, logging and error handling

You produce: a FastAPI model service and a Streamlit demo.

Serving models reliably under real load.

Topics covered

• Model serving over REST and gRPC, with NVIDIA Triton
• ONNX for cross-framework interoperability
• Kubernetes for containerised ML services
• AWS and GCP deployment on scalable cloud architecture

You produce: a model served on Kubernetes in the cloud.

Making AI systems fair and understandable.

Topics covered

• Bias in training data and models
• Fairness across groups
• Explainable AI (XAI)
• Deepfakes and synthetic media

You produce: a bias and explainability review of a model.

Protecting data and meeting regulation.

Topics covered

• Privacy: anonymisation, encryption and federated learning
• GDPR and data protection
• Governance frameworks: OECD, NIST and the EU AI Act

You produce: a governance checklist for an AI product.