AI Deep Dives

OCR with Custom Vision Model: 71% Improvement

How I combined standard OCR with a custom-trained Vision Model and improved extraction accuracy from 58% to 99%.

Bernhard Gotzendorfer

November 28, 2024

9 min read

#OCR#Vision AI#Custom Training#BuchhaltGenie#Document Processing

Facing similar challenges?

I help businesses solve AI challenges where others give up. From initial concept to production-ready solution.

Share Your Challenge

AI Deep Dives

Fine-Tuning vs. RAG: The Decision That Transformed BuchhaltGenie

How I learned after 100+ prototypes that Fine-Tuning and RAG are not competitors - and why we need both in BuchhaltGenie.

December 1, 2024

8 min read

Fine-TuningRAGLLM

Debugging & Solutions

Veto Agent: Why I Contradict Myself

How a Veto Agent in Session 189 exposed my own inflated progress claims - and why controlled self-contradiction leads to real code quality.

December 7, 2025

10 min read

AI QualityMulti-AgentVETO System

Technical How-Tos

RunPod vs. Local Training: The Real Decision Criteria

When does cloud GPU training on RunPod make sense, when local setup? Real decision criteria from Sophie training - including the 10 euros I wasted on an H100.

December 4, 2024

7 min read

RunPodGPU TrainingMachine Learning

Facing similar challenges?

Related Articles

Fine-Tuning vs. RAG: The Decision That Transformed BuchhaltGenie

Veto Agent: Why I Contradict Myself

RunPod vs. Local Training: The Real Decision Criteria