Updated July 2026

AI and Modern Machine Learning for Software Developers

Class Duration

35 hours of live training delivered over 5 days.

Student Prerequisites

Working experience with Python (writing functions, using packages, working with data structures)
Familiarity with basic data concepts such as tables, CSV files, and JSON
No prior machine learning experience or advanced mathematics required

Target Audience

Software developers, data analysts, and engineers who want to move beyond calling AI APIs and learn how machine learning actually works: from raw data through trained, evaluated, production-ready models. Equally valuable for technical leads evaluating ML solutions, QA engineers testing ML-powered features, and data professionals transitioning into machine learning engineering roles. This course pairs well with follow-on training in LLM application development and AI-assisted software engineering, and teams ready for production-scale ML should continue to Advanced Machine Learning and Data Engineering.

Description

This hands-on course walks participants through the complete machine learning lifecycle using the tools that dominate the industry today: pandas and Polars for data preparation, scikit-learn for classical modeling, XGBoost and LightGBM for gradient boosting, and PyTorch for an introduction to deep learning and transfer learning with pretrained networks. Participants work with realistic, messy datasets throughout, cleaning and exploring data, engineering features, training and tuning models, evaluating them honestly, and finally integrating them into real applications with FastAPI and ONNX. All hands-on work is done in Google Colab with agentic AI coding assistance integrated throughout: participants use AI agents to generate, explain, debug, and refactor code the way modern ML teams actually work, while building the judgment to evaluate what the agents produce. The emphasis is relentlessly practical: every concept is applied immediately against scenarios drawn from real engineering work, and participants leave with working code they can adapt to their own projects.

Learning Outcomes

Explain the machine learning workflow from raw data to deployed model, and when to train versus use a pretrained model or hosted API.
Prepare real-world datasets for training: cleaning, encoding, scaling, and splitting with pandas and scikit-learn.
Apply feature engineering techniques and build leakage-free preprocessing pipelines.
Build classification and regression models with scikit-learn and gradient boosting libraries.
Evaluate models with appropriate metrics, cross-validation, and diagnostic tools.
Identify overfitting and tune hyperparameters systematically, tracking experiments with MLflow.
Describe how neural networks learn, and build and train deep learning models with PyTorch.
Apply transfer learning with pretrained networks for image and embedding tasks.
Integrate trained models into applications through model persistence, ONNX export, and FastAPI-based serving.
Use agentic AI coding tools effectively across the ML workflow: generating data preparation code, diagnosing training issues, and reviewing model code.

Training Materials

Comprehensive courseware is distributed online at the start of class. All students receive a downloadable MP4 recording of the training.

Software Requirements

All hands-on work runs in the free tier of Google Colab, which includes built-in agentic AI coding assistance and free GPU access for the deep learning portions. Students need only a Google account and a modern web browser: no local installation required.

Training Topics

The Modern Machine Learning Landscape

AI, machine learning, and deep learning distinctions
Supervised, unsupervised, and reinforcement paradigms
Python ML ecosystem: NumPy, pandas, scikit-learn, PyTorch
Custom models vs. pretrained models vs. hosted APIs
Agentic AI coding assistance across the model lifecycle

Exploring and Understanding Data

Data loading from files, databases, and APIs
Exploratory data analysis with pandas and Polars
Distributions, correlations, and target relationships
Interactive visualization with Plotly
Data quality assessment: missing values, outliers, duplicates

Preparing Training Data

Cleaning and imputation strategies
Encoding categorical variables
Feature scaling and normalization
Train, validation, and test splitting
Class imbalance and resampling techniques

Feature Engineering and Pipelines

Feature selection and importance measures
Derived, datetime, and text features
Dimensionality reduction with PCA
scikit-learn pipelines and column transformers
Data leakage causes and prevention

Core Supervised Learning

Linear and logistic regression families
Regularization: ridge, lasso, elastic net
Decision trees and k-nearest neighbors
Random forests and support vector machines
Algorithm selection by data size and type

Evaluating Models

Baseline models and sanity checks
Regression metrics: MAE, RMSE, R²
Classification metrics: precision, recall, F1, ROC-AUC
Cross-validation strategies
Error analysis and residual diagnostics

Tuning and Ensembles

Overfitting and underfitting diagnostics
Grid, random, and Bayesian hyperparameter search
XGBoost, LightGBM, and CatBoost
Model explainability with SHAP
Experiment tracking with MLflow

Unsupervised Learning

k-means, DBSCAN, and hierarchical clustering
Cluster evaluation and selection
Anomaly and outlier detection
UMAP and t-SNE for visualization

Neural Networks with PyTorch

Tensors, autograd, and computational graphs
Network architecture: layers, activations, loss functions
Training loops, optimizers, and regularization
Debugging training: learning curves and failure modes
GPU acceleration in Colab

Applied Deep Learning and Pretrained Networks

Convolutional networks for image tasks
Transfer learning with pretrained torchvision models
Fine-tuning vs. feature extraction
Embeddings for search and similarity

Shipping Models: Integration and Serving

Model persistence and versioning
ONNX export and inference from C#, Java, and JavaScript
Model serving with FastAPI
Batch vs. real-time inference patterns
Monitoring, drift, and retraining triggers