Clario

Inspiration

Medical imaging reports are packed with technical jargon that most patients can’t understand. You might get an X-ray labeled with nothing special, but still feel short of breath and have no idea what to do next. I built Clario to close that gap: to make radiology results clear, human, and actionable.

What it does

Clario turns complex scan data into patient-friendly reports in under two minutes. You simply upload a chest X-ray (DICOM, PNG, or JPG), fill out a short intake form, and Clario does the rest:

Translates the medical jargon into everyday language
Explains what each finding means in the context of your symptoms and history
Suggests next steps, like possible follow-ups or screenings

In one demo, a 19-year-old’s chest X-ray appears “normal.” But because Clario knows the patient has shortness of breath and a family history of heart disease, it still recommends a cardiac follow-up. The final report includes a confidence score, highlighted findings, and a downloadable PDF summary.

How we built it

Clario is built on three layers working seamlessly together:

Backend (Python/Flask): Runs Hoppr AI models for scan analysis through a tiered approach, including critical, secondary, and additional findings. It then generates a vision-language narrative and uses GPT to create plain-language explanations, all exposed through a REST API with a demo mode.

Frontend (Next.js/TypeScript): Handles uploads (PNG/JPG → DICOM), previews the converted scans, parses AI responses, and generates a polished client-side PDF with real-time progress tracking.

AI Integration: Combines Hoppr AI for image interpretation with GPT for explanation, returning structured JSON with confidence scores and readable summaries.

Challenges we ran into

Supporting multiple image formats meant managing tricky metadata conversions between PNG/JPG and DICOM. We also had to synchronize asynchronous Python pipelines with React state updates without freezing the UI. Structuring GPT outputs in JSON while keeping them natural was another puzzle.

Accomplishments that we're proud of

Fully functional flow from upload → analysis → downloadable PDF

Flexible image input (DICOM, PNG, JPG)

Transparent AI: confidence levels, severity grading, and plain-language explanations

Reliable parsing and robust handling of imperfect LLM outputs

What we learned

We deepened our understanding of DICOM formats, AI model coordination, and state management across Python and Next.js. We learned how to constrain GPT outputs for both structure and empathy, making explanations accurate yet understandable. And we explored how design and UX shape patient trust in healthcare AI.

What's next for Clario

We’re expanding beyond chest X-rays to CT and MRI, adding PACS integration, multilingual support, and even on-device inference for faster, privacy-preserving analysis. Future versions will include peer-review workflows and dosage calculators, all aimed at one goal: making medical imaging results accessible, actionable, and empathetic for everyone.