ML & AI Meetup Bern @inpeek

Talk 1:
How we built a knowledge graph – Powerful RAG chatbot for structured data (Speaker: Fabian Hüni)

Our product Summarix helps manage and prepare employee resumes, project references, and skills for public tenders and bids. To simplify the search for suitable employees or projects, we wanted to go beyond conventional filters and full-text search and enable natural language queries. Unlike typical RAG approaches based on vector embeddings, our data is stored in a relational database (SAP HANA) and includes not only text but also numerical and rich relational information.
In our presentation, we outlined how we developed our chatbot Summy using a knowledge graph for data representation, large language models for SPARQL query generation, and an agent-based AI workflow that enables dynamic interactions while maintaining controlled autonomy. We also showed how we used the open-source tool Langfuse to set up observability and an evaluation pipeline. Finally, we provided insights into our development process and key findings.
This project was part of SAP’s Early Adopter Care Program for the SAP HANA Knowledge Graph Engine and was featured as a success story on the SAP website.

Talk 2:
Spec-driven Development: How AI changes everything (and nothing) (Speaker: Simon Martinelli)

AI accelerates software development, but we still have the same old problem: code becomes the source of truth.
What if we turned that around?
This presentation showcased a new (old) approach to software development. Describe the requirements and everything else—diagrams, models, code, tests—is generated by AI. If the requirements change, everything is automatically updated. No more outdated documents.
Participants learned how to formulate requirements that AI understands and how to link everything together – from business requirements to working code. A real-world project was used to demonstrate how this works in practice.
The core message of the presentation: It’s not about replacing developers. It’s about letting AI take over the boring tasks so that we can concentrate on what’s important: understanding the business requirements.
Der Kern des Vortrags: Es geht nicht darum, Entwicklende zu ersetzen. Es geht darum, KI die langweiligen Aufgaben übernehmen zu lassen, damit wir uns auf das Wesentliche konzentrieren können: die Geschäftsanforderungen zu verstehen.