Course 2 — Coding with a local LLM

Level · IntermediateDuration · ~3 hoursChapters · 23Stack · Ollama + Python + Java + qwen2.5-coder

What this course is about

Five hands-on demos, each in its own GitHub repository under github.com/gneuroneai:

Demo	Repository	What it teaches
Demo 0	gneuroneai/ollama-demo-0-chat-cli	Minimal CLI chat (~50 lines). Used to study the structure of a chat with an LLM at the API level: a list of role/content messages sent to ollama.Client.chat().
Demo 1	gneuroneai/ollama-demo-1-chat-streamlit	Same chat loop served behind a Streamlit web UI. Introduces a graphical interface without changing the underlying mechanics; includes an expander that exposes the raw message list sent to the model.
Demo 2	gneuroneai/ollama-demo-2-comparator	Three models run in parallel on the same prompt. Allows observation of the effect of model size and family on the response.
Demo 3	gneuroneai/ollama-demo-3-agent-java	Single-file CLI agent (~30 useful lines of agent loop). Step-by-step illustration of tool calling: how a Python function is exposed to the model as an action it can invoke.
Demo 4	gneuroneai/ollama-demo-4-trio-agents-java	Three specialised agents (Generate / Verify / Tests) collaborating in a Streamlit UI, with editable system prompts and JUnit 5 verification. Illustrates how the simple agent loop scales to a complete multi-agent project.

Clone any of them individually — they’re self-contained:

git clone https://github.com/gneuroneai/ollama-demo-3-agent-java.git
cd ollama-demo-3-agent-java
.\start.ps1

This course (its own repo: gneuroneai/ollama-course-llm-local) explains everything between those demos and you. No LangChain. No hidden abstractions. Just Ollama + Python + the official ollama-python SDK.

If, by the end, you can edit a system prompt and watch the behaviour change, the course has done its job.

The 23 chapters

#	Chapter	Duration	Type
01	What is an LLM?	8 min	Theory
02	What an LLM cannot do	6 min	Theory
03	Tool calling: the bridge	10 min	Theory + code
04	Ollama vs LangChain	7 min	Theory
04b	Open-source LLM landscape	10 min	Survey
05a	Local vs commercial models	8 min	Theory
05b	Choosing a local model	8 min	Practice
05c	Annex — Local LLM catalogue	20 min	Reference annex
05d	Annex — Quantization, from scratch	20 min	Reference annex
05e	Annex — CPU, GPU, RAM, VRAM: the simple version	20 min	Reference annex
06	Installing the environment	10 min	Practice
06b	Practice 1 — First contact with ollama run	10 min	Hands-on warm-up
06c	Practice 2 — Slash commands explored	15 min	Hands-on warm-up
06d	Practice 3 — Compare 3 models	15 min	Hands-on warm-up
06e	Practice 4 — Ollama Apps & external agents	10 min	Hands-on warm-up
07	Demo 0: minimal CLI chat	10 min	Guided practice
08	Demo 1: Streamlit chat	10 min	Guided practice
08b	Demo 2: 3-way comparator	10 min	Guided practice
09	Demo 3: a simple agent	15 min	Guided practice
10	Demo 4: three agents	20 min	Guided practice
11	Prompt engineering live	12 min	Guided practice
12	Security: workspace & tools	8 min	Theory
13	Going further	5 min	Closing
13b	Annex — LangChain & LangGraph in depth	25 min	Reference annex
13c	Annex — The agentic AI landscape	20 min	Reference annex
14	Fine-tuning a code model	15 min	Advanced

How to read this course

• Linear (recommended): chapters 01 → 06, then 06b → 06c → 06d → 06e (hands-on with Ollama itself, no Python yet), then 07 → 14.
• Express (just run a demo): 06 → 06b → 06d → 07 → 09 → 10.
• Bare Ollama (no Python at all): 01 → 06 → 06b → 06c → 06d → 06e. Stop here.
• Theory only (no installation): 01 → 04.
• Demos only (assumes installed and warmed up): 07 → 08 → 08b → 09 → 10.
• Advanced (train your own model): 05a → 05b → 09 → 10 → 14.

Start with chapter 01 →