Course Outline
Module 1: Core Python for ML Workflows
• Course kickoff and environment setup
Align objectives and establish a reproducible Python ML workspace
• Python language essentials (fast-track)
Review syntax, control flow, functions, and patterns common in ML codebases
• Data structures for ML
Utilize lists, dictionaries, sets, and tuples for features, labels, and metadata
• Comprehensions and functional tools
Implement transformations using comprehensions and higher-order functions
• Object-oriented Python for ML developers
Explore classes, methods, composition, and practical design decisions
• Dataclasses and lightweight modelling
Use typed containers for configuration, examples, and results
• Decorators and context managers
Implement patterns for timing, caching, logging, and resource-safe execution
• Working with files and paths
Handle datasets robustly and manage serialization formats
• Exceptions and defensive programming
Write ML scripts that fail safely and transparently
• Modules, packages, and project structure
Organize reusable ML codebases effectively
• Typing and code quality
Apply type hints, documentation, and maintain a lint-friendly structure
Module 2: Numerical Python, SciPy and Data Handling
• NumPy foundations for vectorised computing
Perform efficient array operations and code with performance in mind
• Indexing, slicing, broadcasting, and shapes
Manipulate tensors safely and reason about shapes
• Linear algebra essentials with NumPy and SciPy
Execute stable matrix operations and decompositions used in ML
• SciPy deep dive
Cover statistics, optimization, curve fitting, and sparse matrices
• Pandas for tabular ML data
Clean, join, aggregate, and prepare datasets
• scikit-learn deep dive
Master the estimator interface, pipelines, and reproducible workflows
• Visualization essentials
Create diagnostic plots for data exploration and model behavior analysis
Module 3: Programming Patterns for Building ML Applications
• From notebook to maintainable project
Refactor exploratory code into structured packages
• Configuration management
Externalize parameters and implement startup validation
• Logging, warnings, and observability
Use structured logging for debuggable ML systems
• Reusable components with OOP and composition
Design extensible transformers and predictors
• Practical design patterns
Apply Pipeline, Factory, Registry, Strategy, and Adapter patterns
• Data validation and schema checks
Prevent silent data issues
• Performance and profiling
Identify bottlenecks and apply optimization techniques
• Model I/O and inference interfaces
Ensure safe persistence and clean prediction interfaces
• End-to-end mini build
Construct a production-style ML pipeline with configuration and logging
Module 4: Statistical Learning for Tabular, Text and Image
• Evaluation foundations
Manage train/validation splits, honest cross-validation, and business-aligned metrics
• Advanced tabular ML
Utilize regularized GLMs, tree ensembles, and leakage-free preprocessing
• Calibration and uncertainty
Implement Platt scaling, isotonic regression, bootstrap, and conformal prediction
• Classical NLP methods
Explore tokenization trade-offs, TF-IDF, linear models, and Naive Bayes
• Topic modelling
Understand LDA fundamentals and practical limitations
• Classical computer vision
Use HOG, PCA, and feature-based pipelines
• Error analysis
Detect bias, label noise, and spurious correlations
• Hands-on labs
Leakage-proof tabular pipeline
Text baseline comparison and interpretation
Classical vision baseline with structured failure analysis
Module 5: Neural Networks for Tabular, Text and Image
• Training loop mastery
Implement clean PyTorch loops with AMP, clipping, and reproducibility
• Optimization and regularization
Manage initialization, normalization, optimizers, and schedulers
• Mixed precision and scaling
Use gradient accumulation and checkpointing strategies
• Tabular neural networks
Work with categorical embeddings, feature crosses, and ablation studies
• Text neural networks
Utilize embeddings, CNNs, BiLSTM or GRU, and sequence handling
• Vision neural networks
Cover CNN fundamentals and ResNet-style architectures
• Hands-on labs
Reusable training framework
Tabular NN vs boosting comparison
CNN with augmentation and scheduling experiments
Module 6: Advanced Neural Architectures
• Transfer learning strategies
Use freeze/unfreeze patterns and discriminative learning rates
• Transformer architectures for text
Explore self-attention internals and fine-tuning approaches
• Vision backbones and dense prediction
Understand ResNet, EfficientNet, Vision Transformers, and U-Net concepts
• Advanced tabular architectures
Utilize TabTransformer, FT-Transformer, and Deep and Cross networks
• Time series considerations
Handle temporal splits and detect covariate shift
• PEFT and efficiency techniques
Navigate LoRA, distillation, and quantization trade-offs
• Hands-on labs
Fine-tune pretrained text transformer
Fine-tune pretrained vision model
Tabular transformer vs GBDT comparison
Module 7: Generative AI Systems
• Prompting fundamentals
Use structured prompting and control generation
• LLM foundations
Cover tokenization, instruction tuning, and hallucination mitigation
• Retrieval-Augmented Generation
Implement chunking, embeddings, hybrid search, and evaluation metrics
• Fine-tuning strategies
Apply LoRA and QLoRA with data quality controls
• Diffusion models
Grasp latent diffusion intuition and practical adaptation
• Synthetic tabular data
Utilize CTGAN and consider privacy implications
• Hands-on labs
Production-style RAG mini-application
Structured output validation with schema enforcement
Optional diffusion experimentation
Module 8: AI Agents and MCP
• Agent loop design
Implement observe, plan, act, reflect, and persist cycles
• Agent architectures
Explore ReAct, plan-and-execute, and multi-agent coordination
• Memory management
Use episodic, semantic, and scratchpad approaches
• Tool integration and safety
Define tool contracts, implement sandboxing, and defend against prompt injection
• Evaluation frameworks
Use replayable traces, task suites, and regression testing
• MCP and protocol-based interoperability
Design MCP servers with secure tool exposure
• Hands-on labs
Build an agent from scratch
Expose tools via MCP-style server
Create evaluation harness with safety constraints
Requirements
Participants are expected to possess a working knowledge of Python programming.
This course is designed for technical professionals at intermediate to advanced levels.
Testimonials (2)
the ML ecosystem not only MLFlow but Optuna, hyperops, docker , docker-compose
Guillaume GAUTIER - OLEA MEDICAL
Course - MLflow
I enjoyed participating in the Kubeflow training, which was held remotely. This training allowed me to consolidate my knowledge for AWS services, K8s, all the devOps tools around Kubeflow which are the necessary bases to properly tackle the subject. I wanted to thank Malawski Marcin for his patience and professionalism for training and advice on best practices. Malawski approaches the subject from different angles, different deployment tools Ansible, EKS kubectl, Terraform. Now I am definitely convinced that I am going into the right field of application.
