The Secret Behind Most AI Tools: RAG. Alex Kihm Explains It Simply.

Dec 15, 2025

Join Dr. Alex Kihm, founder of POMA AI, a pioneer in legal tech and big-data econometrics, as he unpacks the complexities of AI in enterprise settings. Discover why traditional AI search often fails, and how POMA AI’s method reconstructs structure from unstructured data. Alex tackles the chunking problem, explaining how naive approaches distort meaning, and highlights the importance of context engines over outdated retrieval systems. His insights blend humor with a keen understanding of what AI can truly achieve in the corporate world.

Ask episode

AI Snips

Chapters

Transcript

Episode notes

INSIGHT

RAG Is The Source Of Real Knowledge

Retrieval-Augmented Generation (RAG) is the dominant technique to inject up-to-date knowledge into LLMs.
Alex Kihm says most real-world chatbot interactions rely on RAG rather than retraining models.

INSIGHT

PDFs Are For Printers, Not AI

Corporate documents are mostly PDFs designed for printing, not machine understanding.
Alex Kihm argues that this makes enterprise data unreadable to LLMs without preprocessing.

INSIGHT

Chunking Breaks Document Meaning

Embeddings and vector search work on chunks limited by token windows (~8192 tokens).
Kihm explains chunking is unavoidable but often destroys meaning if done naively.

Get the Snipd Podcast app to discover more snips from this episode

Get the app

In this episode of Beginner’s Guide to AI, we sit down with Alex Kihm, founder of POMA AI, to explore how enterprises can finally make sense of their data. AI search is broken, RAG often fails, and corporate documents are notoriously hard for LLMs to interpret.

Alex explains how POMA AI’s patented method reconstructs structure inside unstructured data, enabling powerful, accurate enterprise search.

You’ll hear how his journey from engineering to legal tech to big-data econometrics led to a breakthrough in information structuring. Alex shares why PDFs confuse AI systems, how chunking destroys meaning, and why context engines will replace classical retrieval systems.

This is a deep, funny, insightful conversation about what AI can and cannot do — and how companies can use it responsibly.

📧💌📧
Tune in to get my thoughts and all episodes, don't forget to subscribe to our Newsletter: beginnersguide.nl
📧💌📧

About Dietmar Fischer

Dietmar is a podcaster and AI marketer from Berlin. If you want to elevate your AI strategy or your digital marketing, feel free to reach out anytime at Argoberlin.com

Quotes from the Episode

“Chunking is like reading wrongly sorted text messages from the 90s.”
“Intelligence is pattern recognition — and most enterprise data is not recognisable to machines.”
“PDF was made for printers, not for AI.”
“POMA AI restores the spatial awareness inside documents — the missing context that LLMs need.”
“We don’t do RAG anymore. We build context engines.”
“If your AI breaks the world, show me the invoice.”

Chapters

00:00 Welcome and Introduction

02:45 Alex Kihm’s Background: Engineering, Legal Tech and Early AI Work

10:32 The Problem with RAG, Training, Fine-Tuning and Hallucinations

18:55 The Birth of POMA AI and Solving the Chunking Problem

32:40 How POMA AI Rebuilds Document Structure and Enables True Enterprise Search

45:50 AI Safety, Manipulation Bots and The Future of AI in Business

52:10 Where to Find Alex Kihm and Closing Thoughts

Where to Find the Dr. Alex Kihm

All you need to know about chunking strategies, you'll find here: poma-ai.com
Contact Alex on LinkedIn!

Music credit: "Modern Situations" by Unicorn Heads

Hosted on Acast. See acast.com/privacy for more information.