Clara

Inspiration:

Physicians spend way too much time on intake paperwork during visits. Most of the appointment time is just asking the same questions they've asked a hundred times. Meanwhile, actual patient care is only a small fraction of the visit. It's unproductive for doctors to spend time on forms when they could actually be talking to patients.

What if intake happened before the visit? Let the patient call in the night before, tell their story once, and the doctor walks in already knowing what's going on. This would make every appointment quicker and more effective for both the doctor and the patient.

What it does: Clara is a pre-visit AI intake system that helps doctors prepare more efficiently. Here's the basic idea:

Patient gets a call 24 hours before their appointment
AI (Clara) asks them targeted questions about their symptoms
The system pulls out key metrics such as meds, allergies, red flags, and more!
The doctor sees a clean summary on their dashboard right before the visit.
The doctor can confirm what they see with the patient and will have time to talk about what actually matters instead of wasting time typing at a screen.
Clara integrates seamlessly with EPIC (the hospital patient record software everyone uses), so it's not a hard-to-use, separate tool. The data just flows into the system that doctors already know and love.

We have easy, impactful, and unique onboarding:

Nobody's doing pre-visit intake with AI
Doctors don't have to retrain on new software
It actually handles the boring part, so they can focus on the human part

How we built it:

Design (Figma): We started by drawing out what a physician's day looks like and understanding exactly where they lose time and where they get frustrated. Then we sketched Clara's experience in Figma:

Patient side: Simple phone flow (one question at a time, clear audio feedback, shows estimated call length) Doctor side: Dashboard that takes 5 min to scan (shows meds, symptoms, any red flags) Integration design: Made sure the output looks familiar (SOAP note format + Epic) so doctors don't have to learn anything new We iterated on the dashboard a few times based on feedback before we even started coding.

Tech: Voice: Twilio (handles the phone calls) AI chatbot: Claude + GPT-4o Mini (asks smart follow-up questions, adapts to what you're saying) Data parsing: LLM extracts symptoms/meds/etc from the transcript and structures it Alert system: MD file with rules + AI to catch important stuff (like "chest pain + shortness of breath" = might be cardiac) EHR connection: FHIR API to EPIC (sends data directly into the hospital system) Frontend: HTML + Typescript/React dashboard

Flow: patient calls → transcript gets recorded → AI extracts important info → sends structured note to Epic → doctor sees dashboard. That's it.

Challenges we ran into:

Speech-to-text is terrible at medical words (ex, "Lasix" becomes "lasics"). We added a layer that cleans this up using an LLM with a large medical context value.
Lag time between patient response and GPT thinking time. We would solve this by adding a filtering system that compacts the conversation to mitigate this issue.
To ensure the reliability of AI outputs if used industry-wide, we created a diagnostic reasoning chain that maps out the AI thinking process and provides sources for doctors to verify the output.

Accomplishments that we're proud of:

The voice call works and sounds natural. Check our demo video.
Our UI/UX is highly cohesive with current healthcare systems (Epic) while maintaining a modern aesthetic. This helps doctors minimize onboarding time, ensuring doctors immediately start saving time when using our tool.
We called a few doctors (family connections on our team), pitched the idea, and asked for real advice. - We received positive responses from Haroon Hyder, MD, at Bon Mercy in New York, who helped us understand why patient experience is so valuable and how our idea could significantly impact hospitals.
We also ensured this works in every language by leveraging GPT-4's multilingual capabilities.

What we learned:

Pre-visit is the gap nobody's filling; there's actual business potential in this area of enterprise healthcare.
Talking to Dr. Hyder and showing him a quick Figma mockup further validates our idea. We even iterated on our dashboard multiple times in response to feedback.
Complicated integration is the real blocker. We didn't create a whole new system; we iterated and improved on the software used by 42% of the EHR market.

Market & Strategy:

The healthcare intake market is massive and underserved. Hospitals desperately need intake efficiency solutions, and 42% already use Epic, so our integration path is clear. Competitors like Nuance and Notable Health focus on during-visit documentation. Pre-visit AI intake is uncontested. Our value proposition: doctors save 4-5 hours per clinic per week, patients receive better care, and hospitals reduce no-shows.

What's next for Clara:

Success Metrics: We'll measure physician adoption rate, time saved per appointment, patient completion rate, clinical workflow integration speed, and whether clinics continue using Clara after the pilot.

We'd want to test this with actual clinics to see if it actually saves time like we think it will. Right now, it's a prototype that works: next would be putting it in front of real doctors and seeing if they'd actually use it.

Physician adoption rate (% of doctors using daily)
Time saved per appointment
Transcription accuracy on medical terminology
FHIR integration stability

Future Roadmap:

Phase 1: Prototype Validation (Months 1–3) Deploy with 2–3 pilot clinics. Measure: call completion rate, transcription accuracy, and physician dashboard usability. Fix speech-to-text edge cases and lag time issues based on real usage.

Phase 2: Production Hardening (Months 4–6) Build the filtering system to compact conversations (solve lag time). Implement the diagnostic reasoning chain to show AI thinking. Add HIPAA audit logging. Stress-test FHIR integration with larger patient volumes.

Phase 3: Scaling Foundation (Months 7–9) Create specialty-specific intake templates (e.g., cardio, GI). Build a case-matching feature using vector embeddings. Expand to 3–5 additional clinics, each with a different EHR system, if possible.

Phase 4: Multi-EHR Support (Months 10+) Abstract Epic integration into reusable FHIR modules. Test with various hospitals. Build onboarding automation for new hospital systems.