Beyond OCR: AI Unlocks Document Structure and Vision
Related Episodes
AI Landscape Evolves -- Competition, Copyright, and Compute Bottlenecks
AI's rapid evolution intensifies hardware competition and compute resource battles, while legal battles over copyright and AI companionship reshape its ethical and creative landscape.
View Episode Notes →
2026 AI Landscape: Strategic Shifts, Market Challenges, and Political Narratives
Anthropic leads in coding models, driving Microsoft integration. OpenAI faces distraction, Grok needs differentiation, and Chinese models gain share, reshaping the 2026 AI landscape.
View Episode Notes →
Agentic Engineering Mirrors First-Principles Medical Innovation
Agentic coding platforms achieve team-of-seven output speeds by parallelizing tasks with AI agents, democratizing software creation and accelerating development cycles.
View Episode Notes →
AI Agents Accelerate Development but Require Expertise and Infrastructure
AI agents now code complex projects in minutes, multiplying development speed, but success hinges on specialized prompting and domain expertise.
View Episode Notes →
AI Ecosystem Heats Up: Competition, Partnerships, and Foundational Research
OpenAI's GPT-5.2 advances reasoning and multimodal capabilities, aiming to regain market leadership, while Disney invests $1 billion, signaling a new era for AI in content creation and intellectual property.
View Episode Notes →
ChatGPT as AI Life Partner for ADHD Management and Entrepreneurship
ChatGPT becomes a life partner, optimizing routines, offering emotional support, and transforming rideshare interactions into public AI demonstrations.
View Episode Notes →Resources
Resources & Recommendations
Tools & Software
- Tesseract - A classic OCR tool mentioned as an example of typical OCR models.
- Paddle OCR - Another example of a typical OCR model.
- Docling (IBM) - A toolkit and concept for document structure models, used to predict the structure of a document rather than just extracting text.
- Marked-down (Microsoft) - A toolkit similar to Docling, used in retrieval augmented generation (RAG) systems to preserve document structure.
- Quinn 25 Vision Language Model - A specific language vision model that the speakers have used, noted for being a good model.
- DeepSeek OCR - A newer model that processes documents by splitting them into high-resolution image tokens combined with a global full-resolution view, aiming to preserve more detail than traditional vision language models.
Organizations & Institutions
- IBM - Developed the Docling toolkit for document processing.
- Hugging Face - Released a smaller Docling model suitable for constrained environments.
- Microsoft - Developed the Marked-down toolkit, used in RAG systems.
Websites & Online Resources
- practicalai.fm - The podcast's official website.
- LinkedIn - A platform to connect with the podcast for updates and insights.
- X - A platform to connect with the podcast for updates and insights.
- Blue Sky - A platform to connect with the podcast for updates and insights.
- predictionguard.com - Website for Prediction Guard, a partner providing operational support for the show.