[HYBRID] AI in Software development lifecycle

We're happy to share an upcoming Meetup organized by our friends over at the AI-Innovators Meetup. Looking forward to seeing you there or online. Details below:

Session 1: The future of MLOps and LLMOps in the age of LLMs
In this talk, we introduce a unified approach to building ML systems and AI systems based on feature, training, and inference pipelines.
Whether you are building batch ML systems, real-time ML systems, or agents - all such AI systems can be decomposed into collections of ML pipelines that create features/embedding/context, train or fine-tune models, or make predictions.
In particular, we will look at how to build and operate ML and AI systems with the help of offline testing and online monitoring. For offline tests, we will look at testing features, ML pipeline integration tests, model validation tests, backtests, and blue/green tests. For online monitoring, we will look at observability - metrics and logging, and how to build feature/model monitoring, data validation, and A/B tests. We will also look at evals for LLMs and their key role in validating agentic workflows.Dr. Jim Dowling is the CEO and a co-founder of Hopsworks. He has previously worked at MySQL and as an Associate Prof at KTH Stockholm. Jim organizes the annual feature store summit and is a co-organizer of PyData Stockholm. Jim has written a book for O'Reilly called "Building ML systems with a feature store: batch, real-time, and LLM systems".
#### Session 2: Beyond Vectors—Hybrid + Knowledge Graph RAG for GDPR
We present a stepwise approach to optimizing search over GDPR compliance documents. Starting from a vectors-only RAG baseline, we add hybrid retrieval (keyword + embeddings with LLM re-ranking) and then integrate a Neo4j knowledge graph that encodes GDPR hierarchy, definitions, and cross-references. This progression captures both semantic similarity and explicit legal relationships, improving recall, precision, and answer completeness—especially for citations, acronyms, and multi-hop questions across articles and recitals.
Key takeaways

• Vectors-only RAG misses exact citations and linked provisions, reducing recall on legal queries.
• Hybrid retrieval (lexical + vector) with LLM re-rank recovers acronyms, exact terms, and references more reliably.
• A Neo4j knowledge graph injects structure (articles, recitals, definitions, cross-links) to resolve multi-step queries.
• The graph-augmented hybrid pipeline yields the most consistent gains in precision and completeness for compliance Q&A.
• Pattern generalizes beyond GDPR to other regulations where hierarchical structure and citations matter.

### Event Details:

• Date: 17.09.2025
• Time: 18:00
• Location: Luise-Ullrich-Straße 14, 80336 München

Note: The Event will also happen online. Those who will not be able to join us in person can join under the link:
https://tinyurl.com/2rw7rn84
We’re really excited about this event and can’t wait to share these insights with you. Whether you’re a developer, project manager, or a tech enthusiast, there’s something here for everyone. Come join us and learn how the latest advancements in AI can make your work easier and more efficient.
Looking forward to seeing you there!
Cold beers and delicious food are waiting for you!
The event is co-sponsored by Reply and Hopswork
Please note, we will start the online session at 18:15.