ALL SPEAKERS
AT A GLANCE
Role
VP Engineering and architecture
Organisation
Evana AG
Category
Community
ABOUT THE SPEAKER
Marc Arndt is VP of Engineering and Architecture at Evana AG. With 20+ years in software development and eight years of Go and Kubernetes, he builds production agentic AI for enterprise document processing — and talks about architecture, testing, compliance, and the hard-won lessons of shipping AI with an existing engineering team.
TALK
Documents Are Harder Than Models
Everyone obsesses over which model to use. Then the first real customer PDF arrives — scanned at an angle, in three languages, with tables that carry all the meaning — and the demo falls apart. This session is a deep dive into the part of enterprise RAG nobody puts on slides: document ingestion. We’ll get concrete about OCR quality and how it silently ruins retrieval, layout semantics (why reading order and headings matter more than your embedding model), chunking strategies for real-world files instead of clean blog posts, and handling tables, forms, and multilingual corpora. On the retrieval side we’ll go beyond naive vector search into hybrid approaches that hold up in production. And because enterprise users need to trust answers, we’ll close with verifiable citations: making every answer traceable to the exact passage a human can check. Based on lessons from processing large volumes of real documents in production.
MORE VOICES
Other speakers at code.talks
See them live — and 100+ more.
One ticket, every track. Nov 4–5, 2026 at Kinopolis Hamburg.
Get your ticket





