Self-Hosted LLMs in Practice

Choosing a Model, Running It Locally, and Measuring Whether It Fits.

“Local” is not a magic label. Teams often assume a self-hosted model is automatically more private, cheaper, faster, or production-ready. This course teaches you how to test those assumptions for a defined use case — and how to recommend local, hosted, hybrid, or “do not deploy yet” with evidence.

You will work through a fictional support-ticket scenario with synthetic data. Optional desktop experiments are welcome; the written exercises stand alone if you cannot run a model on your machine.

Who it is for

Product-minded engineers, IT specialists, AI project owners, technical founders, and operations professionals who need to assess or use local language models. Familiarity with AI chat tools and basic concepts such as files, memory, and network access is enough. No programming is required.

What you will leave with

• A clear workload definition and success criteria before model shopping
• A shortlist method that prefers task evidence over leaderboard scores
• A practical view of memory, quantization, formats, and licenses
• A path from a desktop chat experiment to a service boundary
• A quality and speed measurement habit, plus a privacy and data-path checklist
• A one-page deployment decision memo you can defend

How the course works

Five modules, taken sequentially. Each topic ends with a short check; each module has a checkpoint quiz. Keep answers to the in-lesson worksheets in one document — that becomes your final decision dossier.

Do not use private customer records, confidential documents, or real credentials in any experiment unless your organization has explicitly approved the environment and controls.