Anders Ottesen & Anton Baht

Challenges and Use Cases in Processing 11 Million Documents for Agentic AI

Over the past year, PFA’s AI team has built a large‑scale document processing pipeline that turns 11 million heterogeneous, messy documents into the foundation for a production‑ready agentic AI assistant for caseworkers. This talk walks through the end‑to‑end journey from raw, unstructured data to searchable, structured knowledge that an AI agent can reliably act on. We dive into the key challenges: securing business buy‑in, designing the right infrastructure, processing data at this scale, building a robust agent on top, raising AI literacy among users, and ensuring governance at every step. Finally, we share concrete lessons learned from running a large‑scale GenAI project in a highly regulated enterprise: what failed in early iterations, what worked, and how we operate and evolve the solution today.

Anders Ottsen is a Senior Machine Learning Engineer at PFA. He has primarily worked on Data & GenAI infrastructure as well as products that utilizes it.

Anton Baht is a Senior Data Scientist. He has primarily worked on GenAI products.