When Trust Is Everything: Building AI for Physicians at Healio

This episode of Just Now Possible features three leaders from Healio, a 125-year-old medical publisher, discussing how they built Healio AI, an assistant helping physicians prepare for patient care. The core argument is that in high-stakes domains, trust, accuracy, and transparency must be designed in from the start rather than added later. The team describes learning from a survey of 300 healthcare professionals, discovering that physicians wanted help with patient communication more than diagnostics, and prototyping quickly before iterating. They explain the technical approach of hybrid retrieval across trusted sources, citation design that builds credibility, privacy guardrails, and an evaluation system combining LLM judges with physician feedback. The lessons matter well beyond healthcare for anyone building AI where errors carry real cost.

01Key takeaways

• In high-stakes domains, earn trust through verifiable sources and visible citations before asking users to rely on outputs.
• Validate your assumptions about user needs with research; the most valuable use case may differ from what you expected.
• Build quick prototypes to learn fast, even if the first version is a weekend hack.
• Combine multiple retrieval methods, such as lexical and semantic search, to improve answer quality across trusted sources.
• Use LLM-as-judge evals to scale quality checks, but keep domain experts in the loop as the ultimate source of truth.
• Mask personal health information at the input stage and design compliance in from the start.

02Key sections

Discovering the real need

The team surveyed hundreds of healthcare professionals and found that their assumptions were off. Physicians asked for help with patient communication and empathy, not only clinical answers.

Prototyping fast

After starting with design mockups, the team built a working prototype over a weekend, using the prototype as a way to learn rather than as a finished product.

Retrieval across trusted sources

Healio combines lexical, vector, and semantic search over several authoritative sources, and the episode highlights that even accessing PubMed involves multiple trade-offs.

Designing for trust

Citations are presented with subscripts, hover states, and progressive disclosure so physicians can verify claims without being overwhelmed.

Evaluating quality

A set of eight LLM judges scores responses on dimensions like safety and medical accuracy, but physician feedback is treated as the final authority in high-stakes contexts.

03From the post

“Listen to this episode on: Spotify | Apple Podcasts How do you build an AI product for an audience that can't afford to be wrong—and won't trust you until you prove it? In this episode of Just Now Possible, Teresa Torres talks with three leaders from Healio—Jennifer”

04Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.