SpeedChart.ai
From Startup Idea to Research Questions
Co-Founder & Chief Operating Officer
Eric Douangdara, Sarayah B Obonyo, Connie Huang, Skyler Chen, & Shelly Zhao
Rethinking the Clinical Conversation
SpeedChart.ai began as a graduate entrepreneurship project exploring how artificial intelligence could help patients better understand and remember their medical visits. Our team designed a concept that used real-time speech-to-text, AI-generated visit summaries, and personalized task checklists to make healthcare information more accessible and significantly easier to act on.
As I worked on the project, however, I found myself becoming less interested in the technology itself and deeply drawn to the complex information ecosystem surrounding it. Building SpeedChart.ai raised structural questions that extended far beyond product design:
Information Governance: Who owns and governs AI-generated clinical information?
Data Privacy: How should sensitive patient conversations be stored, secured, or permanently deleted?
Linguistic Representation: What happens when automated speech recognition fails to comprehend patients with diverse accents or multilingual speech?
Clinical Accuracy: How do transcription errors alter diagnoses, treatment paths, and patient trust?
System Accountability: Who is responsible when artificial intelligence misrepresents what a patient actually said during a vulnerable clinical encounter?
Algorithmic Equity: How can AI improve access to healthcare without introducing entirely new disparities?
These questions fundamentally changed the direction of my career. Rather than focusing solely on building software applications, I became increasingly committed to patient-centered research, health information quality, AI governance, and the policies that determine whether these technologies improve care equitably. Today, I see SpeedChart.ai not simply as a startup concept, but as the project that sparked my true research agenda.
Linguistic Diversity and Accent Bias
As someone who grew up outside the United States, I became particularly interested in how speech recognition systems handle linguistic diversity. If an AI system misunderstands a patient's accent, ignores code-switching, or incorrectly transcribes culturally specific terms, the resulting record may no longer reflect what the patient actually said. For me, that raises urgent questions not just about model accuracy, but about equity, information quality, and patient safety.
This project taught me that designing AI for healthcare isn't only a technical challenge — it is an information challenge. Building trustworthy systems requires understanding not only what AI can do, but how information is created, interpreted, governed, and experienced by real patients. Those questions continue to shape my work today.