AI Breakfast #26
Executive Summary
At our AI breakfast meetup, a group of professionals from diverse backgrounds discussed topics ranging from AI gaps in healthcare to language barriers in portfolio management. The attendees also shared insights on their projects, including healthcare data quality improvements and WeChat automation solutions.
Member Work
Healthcare Data Quality and Export Opportunities
A participant from the medical industry in Shanghai highlighted challenges with healthcare data quality, which limits AI implementation. They discussed the potential of exporting Chinese hospital management AI systems globally, emphasizing efficiency gains and cost advantages over Western systems.
Language Barriers and Cognitive Skills
A portfolio manager shared their struggle with bridging language barriers between English and Chinese. They expressed concerns about AI dependency leading to cognitive skill atrophy and emphasized the need for AI to reinforce long-term memory rather than replace it.
WeChat Automation and Phone Farm Solutions
An entrepreneur and consultant in XR/AI/VR discussed the challenge of integrating AI with non-code communication channels, specifically WeChat. They explored solutions like setting up phone farms in Australia and the US to bypass VPN and geo-blocking restrictions.
Group Discussions
Technical Solutions
The group explored coding workflow optimization, including the use of skills concepts for better context management and the hierarchical structure of AgentsMD for team collaboration. They also discussed Prodigy AI for custom model training and spec-driven development approaches.
Healthcare AI Export
The conversation touched on the opportunity to export Chinese hospital systems, which offer significant cost savings and efficiency gains compared to US systems. The discussion included regulatory advantages and challenges in technology transfer and local adaptation.
AI Development Insights
Participants shared insights on context management, team collaboration, and custom training for domain-specific models. They also discussed pose detection applications using Apple's built-in machine learning frameworks.
Other Resources
- Granola: A bot-free meeting transcription tool that listens to device audio. Attendees found it useful for quickly generating meeting summaries.
- Prodigy: An annotation tool for building custom AI models with active learning. Some members used it for domain-specific models like physio posture correction.
- Zed: A high-performance code editor built in Rust. Mentioned alongside Cursor and Claude Code for coding workflow optimization.
- AGENTS.md: An open standard for configuring AI agent behavior via Markdown files. Used for consistent AI behavior across projects.
- Apple Vision Framework: Apple's framework for body pose detection. Participants discussed using it for movement analysis applications. (Note: This was referred to as "MLKit" in conversation, though MLKit is actually Google's SDK; Apple's equivalent is the Vision framework with Core ML.)