Introduction to Machine Learning and AI Engineering

Many IT organizations are adopting LLMs faster than their teams understand them. That gap brings real risk: hallucinated answers in internal tools, sensitive data entered into personal AI accounts, and AI-generated code that nobody on the team can explain or secure. This course builds a shared foundation so developers and engineers treat foundation models as powerful systems that still need careful engineering.

Role fit: This course suits programmers, developers, data analysts, DevOps engineers, and IT practitioners who already have basic programming experience. It does not teach software engineering itself. Team Leads can assign it to engineers moving into AI application developer, ML engineer, or AI solutions engineer roles.

Time investment: Plan for about 27 hours across 30 skills per learner. The first half goes deep on ML fundamentals in PyTorch. TensorFlow, NLP, and AI engineering topics such as prompt engineering, RAG, evaluation, tools, and agents are covered at an introductory level. The course discusses deployment but does not perform it.

Risk and governance: The course covers practices that support responsible AI use, including:

• Keeping PII and protected health information out of unmanaged AI accounts
• Knowing which libraries AI tools import, to reduce security exposure
• Designing RAG systems that answer "not found" instead of guessing
• Setting agent rules such as "do not invent policies" and escalating when needed

Change management: Learners code along in Colab notebooks and complete a challenge at the end of each skill. These checkpoints confirm understanding before learners move on. Use Playlists to sequence this course with leadership-focused AI training for a stakeholder-aligned rollout, and use Team Reporting to track progress across the team.

IT teams use this training so AI work moves from experiments to grounded, testable applications. The course uses realistic scenarios: a fleet of self-driving taxis generating messy trip data with missing timestamps and negative fares, a support assistant that has to answer refund-policy questions, and an agent that escalates instead of inventing answers.

• Fewer hallucinated answers reaching users: Your team builds a RAG application with Chroma and LangChain. It grounds responses in a knowledge base drawn from PDFs and internal policy documents. When an answer isn't in the retrieved context, the system is taught to return "not found" rather than guess.
• Repeatable quality checks before release: Practitioners write code-based evaluators and automate evaluation workflows. These check not-found behavior and required source citations, so AI output quality is measured rather than assumed.
• Faster diagnosis of AI application behavior: Learners practice thinking through how AI systems use context, tools, memory, and evaluation. That habit helps teams identify whether a poor result comes from missing context, weak retrieval, unclear instructions, or an agent that needs stricter operating rules.
• Safer AI adoption across the org: Practitioners learn why sensitive data such as PII and PHI should stay out of personal AI accounts and why enterprise accounts matter. They also learn to vet the libraries AI coding tools import and to set operating rules for agents, such as using tools when needed, not inventing policies, and escalating when uncertain.

The result is a team that can explain why models make mistakes, judge when a foundation model needs better context or tools, and build AI features your organization can trust.

Knowledge

• How AI, machine learning, deep learning, and generative AI differ, and how ML work — training data, labels, and evaluation metrics — compares with AI engineering on top of foundation models
• The core building blocks of learning models: linear and logistic regression, the perceptron, and scalars, vectors, matrices, and tensors
• How loss functions, gradients, and optimizers work together, including when to use mean squared error versus cross-entropy loss
• The ML pipeline end to end, from data ingestion and cleanup of messy real-world data through training and evaluation
• Why LLM outputs are probabilistic, how token costs and context shape responses, and why hallucination must always be mitigated
• Data-handling risks with AI tools, such as keeping PII and protected health information out of unmanaged accounts

Ability

• Plan ML workflows in pseudocode before writing code, then build regression models, neural networks, and an image classifier in PyTorch
• Diagnose and fix common tensor shape and data type errors that derail PyTorch training
• Build and improve an NLP text classifier with TensorFlow using an RNN
• Write structured prompts using context, role, expectation, and few-shot examples to get more predictable LLM output
• Build a RAG application with Chroma and LangChain that answers from a knowledge base and returns “not found” when the answer isn’t in the retrieved context
• Automate evaluation of AI outputs and integrate tools that ground responses in an introductory AI engineering workflow