Local AI Hands-On

Target Audience

IT staff who want to operate and develop local AI systems themselves.

Learning Goals

• Setting up a local AI infrastructure
• Hands-on building your own agents
• Discussion of concrete use cases

Topics

Hardware & Model Selection

Running LLMs locally enables development without massive costs and allows processing of sensitive data.

• What do I need to run LLMs locally?
• What hardware is required and what optimization options exist for limited resources?
• How do size-optimized models affect quality and performance?

The goal is to set up your own server and successfully send sample requests to the system. At the same time, an understanding of the impact of optimizations will be developed.

Keywords: administration, LLM-server, GPU parallelization, model formats, quantization, LLM-protocols, hardware limitations, performance, perplexity, KV-cache

Use Cases

Agentic systems are far more than coding assistants. Custom use cases for agentic systems will be introduced and discussed using live examples.

• What is the internal architecture of the harness/scaffolding?
• Which pitfalls have been observed?
• Where are the limits of the respective systems?
• What techniques exist to make agentic systems robust?

The goal is to develop an understanding of how to build or implement own systems and when it makes sense — i.e. which scenarios in your own organization can benefit from the new opportunities agentic systems offer.

Keywords: development, agentic systems, pitfalls, use cases, best practices

Custom Agentic Harness — Hands-On Session

LLMs are not intelligent systems but interpreters of non-deterministic programming languages. The prompt is the program. This may sound complicated, but is easy to understand with an example. We want to convey this mental paradigm shift here — moving from "magical black boxes" to "predictable machines". At the same time, it becomes clear which methods are necessary to set up, test, and maintain such systems sustainably.

To this end, we will develop a minimal example of an agentic system live and deepen the understanding of previously learned properties using the emerging example.

Keywords: prompt, bias, priming, emotion, non-deterministic logic, reasoning, KV-cache

Miscellaneous & Interactive Discussion

The workshop spans only two days and can therefore only partially cover the functionality, limits, and possibilities of agentic systems. This session provides space for specific topics and concrete questions that arose during the workshop.

The space can also be used for a joint brainstorming session on possible use cases. There are no specific topics or content restrictions here. We will define these on-site based on your needs.

Keywords: brainstorming, questions, discussion, understanding, future and vision

Format

This workshop is typically conducted on-site as a full-day event.