HandsOnAI

Overview

HandsOnAI is a modular Python toolkit that teaches three core AI interaction patterns through progressive, hands-on learning. Students build and interact with each pattern directly, moving from simple chat to retrieval-augmented generation to autonomous agents.

GitHub: teaching-repositories/hands-on-ai Install: pip install hands-on-ai

Three Modules, Three Patterns

1. Chat — Prompt Engineering

The simplest pattern: send a message, get a response. Students learn how system prompts, temperature, and persona design shape model behaviour.

chat --personality teacher "Explain recursion"

Multiple built-in personalities demonstrate how the same model produces radically different responses depending on the system prompt.

2. RAG — Retrieval-Augmented Generation

Students load documents and ask questions grounded in that content. The module handles chunking, embedding, similarity search, and context injection — but exposes each step so students understand what RAG actually does.

rag load report.pdf
rag ask "What were the key findings?"

This is where students learn that LLMs don’t know things — they generate responses based on context. Providing the right context is the skill.

3. Agent — ReAct Reasoning

A ReAct-style agent that reasons step by step, selects tools, and composes multi-step answers. Students see the reasoning loop in action: think, act, observe, repeat.

agent "What's the weather in Perth and should I bring an umbrella?"

The agent module teaches the pattern behind every AI agent framework — the same loop that powers LangChain, CrewAI, and AutoGen, stripped to its essentials.

Provider Agnostic

HandsOnAI works with any OpenAI-compatible API:

• Local: Ollama, vLLM, LocalAI
• Cloud: OpenAI, Google Gemini, Groq, Together AI, OpenRouter
• Hugging Face: Inference API endpoints

Students learn the industry-standard API pattern once and apply it everywhere. No vendor lock-in.

The Rehearsal

HandsOnAI is the intermediate step in the AI development track. Students who have mastered basic LLM interaction with SimpleBot move here to learn the three patterns that underpin most real-world AI applications. The progression — chat → RAG → agents — mirrors how production AI systems are actually built.

Each module is self-contained. An educator can teach just chat in a single session, or run the full three-module sequence across a semester.

Part of LocoEnsayo

LocoEnsayo covers two types of rehearsal: professional skills (talking to AI characters) and AI development (learning to build AI applications). HandsOnAI is the intermediate step — after SimpleBot, before building full applications like CloudCore or TalkBuddy.