Goal: run QAZTECH AZ-104 Tutor locally and use a prompt you can verify. Allow 25 minutes of active study, plus the download time required by your connection and device.

Choose the published model

Start with the QAZTECH AZ-104 — Open-source AI Tutor repository. Follow its current distribution links and read the model limitations. The model is a preview learning tool based on an 8B open model; it is not an official Microsoft product. Its coverage and multilingual quality are uneven. Use official documentation for topics that it cannot answer reliably.

Install Ollama using its official instructions for your operating system. In a terminal, confirm the installation and start the model:

ollama --version
ollama run qaztech-platform/az104-tutor

The first run downloads model data. Check free storage and available memory before starting. A quantised file’s size is not the total memory required during inference: the runtime and conversation context also use memory. Begin with a short prompt and close unnecessary applications if the device is constrained. Record the model name and runtime version in your session notes.

Ask for an explanation you can test

I am beginning Azure administration. Explain the difference between an Azure subscription and a resource group. Give one practical example. Ask me one question and wait for my answer. Identify claims I should verify in Microsoft Learn.

Answer the question before asking for an explanation. If an answer seems plausible but uncertain, ask which service behaviour makes it correct and locate the official documentation yourself. Avoid requesting an entire course in one prompt; work with one topic at a time.

Alternative: a GGUF-compatible local runtime

The GGUF model card provides quantised files. Choose a runtime that supports the model architecture, then select a file that fits your device and use the card’s chat-template guidance. The separate Transformers release serves a different runtime path. Do not rename a checkpoint and assume it becomes a compatible GGUF model.

Check the privacy boundary

Confirm that the selected model is local rather than a hosted fallback. Downloads, updates and documentation requests still use the network. Keep the local model service bound to an appropriate local interface; do not expose it to the internet for this exercise. Your surrounding application and its settings also determine where data goes.

Completion evidence: record the runtime version, model identifier, one question you answered and the official source used to verify its explanation. Do not include passwords or tokens in that record.