Cloud Native Computing Rheinland - October Edition @ viadee

Hi everyone! Our next meetup will take place in the viadee offices in Cologne. We are super excited to be back from our summer break and are looking forward to this event with exciting talks.
Please keep in mind that we only have space for a limited number of participants. Please only RSVP if you actually intend on joining for the event, and please cancel your RSVP again should your plans change so others can take your spot. Thanks for your cooperation!
Drinks and snacks will be provided.

Preliminary Schedule:
18:00 Beginning and Socialising
18:30 First Talk and discussion
19:15 Pizza & Drinks
19:45 Second Talk and discussion
20:30 Socialising and further discussions

From cloud to edge : building practical computer vision pipelines with Kubeflow and Ultralytics YOLO models
Our two talks this evening work in conjunction with each other. Both speakers are talking together about the same topic, but each one focuses on different aspects.

Part #1 Matt Chidambaram - Application Manager, Vodafone GA-Online

Kubeflow, a CNCF-incubated project, enables streamlining of ML pipelines on Kubernetes, but many real-world scenarios benefit from integrating external frameworks.
We present an **Edge AI Workflow** that uses an OpenCV / Ultralytics based model ( for computer vision ) which can be trained and fine-tuned, and Kubeflow for orchestration and deployment.
The Kubeflow pipeline targets primarily lightweight edge hardware such as Raspberry Pi but potentially also cloud backends in a distributed, layered setup.
Through this use case, we will aim to highlight some of Kubeflow’s strengths and limitations while showing how complementary tools such as MLFLow can extend its capabilities.

Part #2 - Andreas Badelt - Solution Architect, Freelance

Abstract
Large Language Models and AI Agents based on them have come a long way in recent years. But with all the innovation you also need standards - so that these innovations can spread widely and be used efficiently. Proper standardisation at system borders makes these systems easily interoperable and render all the annoying and costly custom coding obsolete. The relevant system borders or interfaces for LLMs/Agents are obvious: Communication with a client (user or system); communication with tools that support the Agent in fulfilling its task (like doing a web search, accessing a database or CRM system etc.); and communication with other agents in a multi-agent environment. For the latter two, a pair of protocols have materialised over the last year or so: The „Model Context Protocol“ (MCP) initiated by Anthropic, which helps integrate tools of all sorts into LLM-based system, from apps like Claude Desktop to actual AI Agents. And the „Agent2Agent Protocol“ (A2A), started a few years earlier at Google to create - you may have guessed it by now - efficient agent-to-agent communication.

This talk will introduce the two protocols and show their principles and usage alongside a real-life project: How can you best leverage them in the construction of (multi-) agent systems, what are their strengths and weaknesses?

Disclaimer
By attending our meetups you agree that the photos and videos taken during the event can be used in social media platforms like LinkedIn to promote our future events.