AI with local and open models

About this course

This course is about running open models on your own hardware instead of a cloud API: when that's actually worth it, what it takes, and how not to lose quality along the way. It solves the problem of choosing between local and cloud inference based on real constraints, not trends.

The course moves from total cost of ownership and open-weight licenses through quantization and memory bandwidth to configuration choices, deploying a working server, and running it in production. Projects run on your own setup: a hardware profile for your task with measured memory and speed, a working local server brought to a load-ready state, and an evidence base for a build-vs-buy decision on local models.

Built for developers and engineers who already call models through an API and want to understand when and how to move to a local or open alternative. You'll need basic Python and comfort with the command line — no prior GPU knowledge required, the course covers that from the ground up.

What you'll learn

• Calculate total cost of ownership for a local model versus a cloud call
• Read open-weight licenses and understand what they restrict
• Estimate memory footprint, precision, and quantization trade-offs
• Deploy a model as a server process with a network interface
• Tune inference parameters, batching, and context length
• Fairly compare a local model's quality against a cloud model
• Diagnose common local serving failures
• Secure a deployment: isolation, access, and what ends up in logs

What you'll end up with

• A hardware profile with measured memory, speed, and quality for your task
• A deployed local server brought to a load-ready state
• An evidence base comparing your local and cloud model options
• A certificate on paid plans

Not for you if you just want quick API access to any model without asking why you'd run it locally at all.