Tom Kaltofen
Engineer
mloda.ai
Tom Kaltofen is a Berlin-based data and AI engineer and the creator of mloda, an open-source (Apache-2.0) Python framework for declarative, plugin-based data access in AI workflows. After years of building in-house feature stores, ML pipelines, and RAG systems, he kept seeing "model" failures that were really data-context failures, and built mloda to close that gap: the layer between raw sources and the feature stores, RAG frameworks, and agent orchestrators that consume them. His work on open government data is currently funded by the Prototype Fund.
Why AI Agents Fail on Enterprise Data and What Coding Agents Get Right
Session
When AI agents fail on enterprise data, the root cause is usually the data layer, not the model. Coding agents avoid this because their context is deterministic by design. Enterprise data has none of that by default.
This talk introduces the handover layer pattern and demonstrates it live: an AI agent discovering transformations, applying PII redaction, and switching data sources without hardcoded paths.
Takeaways:
- How to separate data context failures from model failures
- A concrete architectural pattern for deterministic data access
- A live demo reproducible locally with open-source tooling