Hands-On Local LLMs: Run and Build with AI on Your Own Hardware - C2
Leverage AI locally to build modern applications quickly, reliably and cost effectively
Downloading a local model and getting it to answer a prompt is easy.
Choosing the right model, running it efficiently on your hardware and integrating it into a reliable application is much harder.
In this live three-hour workshop, you’ll work through the complete local LLM stack, from model and hardware selection to quantization, inference, application integration, evaluation and LoRA.
You’ll explore tools including Ollama, llama.cpp, MLX and vLLM while learning how to make practical decisions around privacy, cost, latency, licensing, performance and reliability.
The goal is to help you confidently answer:
When should we run an LLM locally, what should we run, and how do we make it reliable enough for a real application?
Why Attend?
Local LLMs can keep sensitive data inside your environment, reduce dependence on hosted APIs, support offline applications and give you greater control over cost and infrastructure.
But those advantages depend on choosing the right model, hardware and serving stack.
This workshop will help you move beyond experimentation and understand the engineering decisions behind a dependable local AI system.
What You’ll Learn
- Compare Gemma, Llama, Mistral and Qwen for different use cases
- Estimate RAM and VRAM requirements before running a model
- Understand how quantization affects memory, speed and quality
- Choose between Ollama, llama.cpp, MLX and vLLM
- Connect local models to assistants, RAG systems and applications
- Add sandboxing and guardrails to local agent execution
- Evaluate quality, latency, memory use and regressions
- Decide when to use prompting, RAG or LoRA
By the end, you’ll have a clear decision framework for building local AI systems from hardware and model selection to quantization, inference and application integration.
Meet Your Instructors
Murat Sari
Murat brings hands-on experience from industrial machine vision and local AI systems, where inference must run reliably on-site.
Rainer Hahnekamp
Rainer is a Google Developer Expert, core NgRx maintainer, creator of Sheriff and NgRx Toolkit, and an enterprise software engineer with 15 years of experience.
Together, Murat and Rainer will show you how to make local models work technically and how to integrate them reliably into real software systems.
Who Should Attend?
This workshop is designed for:
- Software developers building AI-enabled applications
- AI engineers comparing open models and inference stacks
- Architects evaluating private or sovereign AI
- Platform and DevOps engineers managing model infrastructure
- Security engineers exploring safer local agent execution
- Teams deciding between local models and hosted APIs
No previous experience with local inference engines or fine-tuning is required.
What’s Included
- Three-hour live, hands-on workshop
- Practical model and hardware selection framework
- Guidance across leading local inference tools
- Live Q&A with both instructors
- Full workshop recording
- Certificate of completion
Good to know
Highlights
- 2 hours
- Online
Refund Policy
Location
Online event
Agenda
Opening Demonstration and Architecture Overview
See a complete local LLM application and trace the path from model files and hardware through the inference engine to the application. This shared architecture becomes the reference point for the rest of the workshop.
Local LLM Landscape and Model Selection
Compare relevant model families, including Gemma, Llama, Mistral, and Qwen. Select models by task, language, context length, tool use, multimodality, license, model size, and hardware requirements.
Hardware Model Formats and Quantization
Cover CPU, GPU, Apple Silicon, RAM and VRAM sizing, as well as model formats such as GGUF and SafeTensors. Understand how precision and quantization choices affect memory consumption, compatibility, speed, and output quality.