How can modern OCR systems combine high extraction quality with efficiency, reliability and maintainability when resources are limited?
In this session, Olmo Barberis shares insights from building OCR solutions for real-world applications and compares traditional and modern approaches to document processing.
The session covers:
- traditional character recognition approaches with dictionaries and rule-based post-processing
- modern OCR solutions using open-source frameworks, vision models and LLMs
- contextual correction to improve extraction quality and robustness
- strengths and limitations of different OCR approaches
- engineering trade-offs under resource constraints
- combining visual and textual signals for structured information extraction
- maintaining a consistent and reliable user experience
The focus is on practical engineering decisions: how to build OCR systems that work reliably in real-world environments without ignoring resource constraints.
Olmo Barberis is a Product Manager and Software Engineer at Karakun AG in Basel. He has experience in automated document processing and semantic search engine development, with a focus on AI technologies.