Short Coding Course — South Africa
LLM APIs, RAG, AI Agents & MCP
Clients and employers want AI inside their products — not just in their developers' editors. This hands-on course teaches Python developers to build reliable AI features with the OpenAI and Claude APIs: structured output, tool calling, retrieval over your own documents, agents and MCP servers — tested with evals and ready for production.
Tokens, context windows, temperature and reasoning models — what actually drives quality, speed and cost, and how to choose the right model for each job.
Messages, multi-turn conversations, streaming, error handling, retries, rate limits and cost tracking — wrapped in a provider-agnostic layer you can reuse.
System prompts, few-shot examples, prompt templates and prompt caching — and how to test prompts so they stay reliable as your app grows.
Getting dependable JSON with schemas and Pydantic, extracting data from documents and emails, and processing large batches cost-effectively.
Letting the model call your own functions and APIs, chaining multiple tool calls, and handling errors and retries safely.
Embeddings, chunking, vector stores, re-ranking and measuring retrieval quality — build a “chat with our company documents” assistant that cites its sources.
When an agent beats a simple workflow, building agents with the OpenAI Agents SDK and Claude Agent SDK, and putting limits on what an agent is allowed to do.
Build your own MCP server to expose your company's data and APIs to Claude and other AI tools — with authentication and least-privilege access.
Running open-weight models with Ollama, keeping sensitive data on your own infrastructure for POPIA, and weighing cost, speed and quality against cloud APIs.
Automated evals, LLM-as-judge, prompt injection, guardrails and monitoring cost in production. Capstone: build and demo an AI feature with tests and evals.
Building with LLMs is for developers who already write Python. Make sure you're ready before enrolling.
Comfortable with functions, classes, packages, pip and virtual environments. You'll write Python in every lab.
Calling APIs, reading JSON responses and handling HTTP errors — LLMs are used through APIs.
Cloning, committing and running scripts from a terminal.
API access: OpenAI and Anthropic API accounts for the labs. We confirm API access options with you when you book.
A capable laptop: 8 GB RAM minimum; 16 GB recommended to run local models with Ollama.
If you're new to Python. Bootcamp graduates can go straight to Building with LLMs.
Ready to take this course! Build AI features with LLM APIs, RAG, agents and MCP.
Go under the hood: how machine learning models learn and how to evaluate them.
You build software in Python and want to add chat, search, data extraction and automation powered by LLMs.
Take the AI module from the Web Developer Bootcamp much further — RAG, agents, MCP and evals.
Build assistants over your company's documents and systems — with control over cost, privacy and quality.
Understand what LLMs can and can't do reliably, what they cost, and when to build versus buy.
All courses are taught by Code College’s senior development trainers — practitioners with real-world industry experience, backed by over 20 years of professional developer training since 2004.
You write code from day one. Every concept is reinforced with real exercises and a practical project you keep.
Join 100% live online from anywhere, or hybrid — one or two days a week at our Midrand campus and the rest live online. Full-time face-to-face attendance can be requested, and is possible when an onsite group class is running.
Receive a Code College certificate on completion — recognised by employers across South Africa.
No. This course is about building applications with existing large language models through their APIs — you won't train models. If you want to learn how machine learning models are built and evaluated, see Python for AI & Machine Learning.
Yes. All labs use Python, the most common language for AI work. Developers who know JavaScript can follow the concepts easily — and Code College Web Developer Bootcamp graduates already know both Python and JavaScript. Java teams can learn the same skills in Spring Boot on our Spring AI course.
The labs use the OpenAI and Claude APIs, which charge per use — a course week typically costs very little. We confirm API access options with you when you book, and many labs also run free on local models with Ollama.
Step by step you build a provider-agnostic API layer, a data-extraction tool, an assistant that calls your own functions, a RAG assistant over your own documents, an agent and an MCP server. The capstone brings them together into an AI feature with tests and evals that you demo on the last day.
AI Tools for Developers teaches you to use AI coding assistants such as Claude Code and GitHub Copilot to write your own code. Building with LLMs teaches you to put AI inside the applications you build for your users.
Individuals join live online from anywhere, or hybrid — one or two days a week at our Midrand campus and the rest live online. Group courses run live online or hybrid, with fully onsite delivery at our Midrand campus optional, on request. Contact us for a group quote.
Yes. Code College is a B-BBEE Level 4 contributor and training spend qualifies as skills development under the B-BBEE scorecard. The course fee also counts toward your SDL (Skills Development Levy) spend.
Book the Building with LLMs course today — full-time (5 days) or part-time. Johannesburg and online.