Care Memory

Inspiration

A consultation is one of the most information-dense conversations a person ever has, and most of it disappears the moment they leave the room.

Doctors see dozens of patients a week. A patient may come back after three days, three weeks or three months, and the doctor has to rebuild the story from short notes and memory. Patients forget too. Memory fades roughly exponentially:

$$R = e^{-t/S}$$

where $R$ is how much is retained, $t$ is time since the conversation, and $S$ is the strength of the memory. A diagnosis, a dose change or a warning sign can fade within days. When that happens, details get lost and miscommunication follows.

We asked a simple question: what if every conversation could be continued, instead of restarted?

That became Care Memory.

What it does

Care Memory gives doctors and patients a shared, readable record of every consultation.

• Doctors sign in, open a dashboard of their patients (total patients, total sessions, this week, today), and start a new session by entering a patient ID and then either uploading an audio file or recording the conversation live.
• Each recording is turned into a clean Markdown session note covering symptoms, history, findings, diagnosis, medication changes, advice, tests ordered and follow-up date.
• Patients enter their ID and see their sessions in a calm reading view, with the note in the center and a timeline of previous sessions on the left.
• Admins have a separate panel for adding doctors.

Before a follow-up, a doctor can read what was said last time in seconds, so they don't depend on memory.

How we built it

• Frontend: React 18 with Vite, React Router, Tailwind CSS, and Framer Motion for slow, quiet animations.
• Audio: the browser MediaRecorder API for live recording, wavesurfer.js for waveforms, and file upload for existing recordings.
• Reading view: react-markdown with remark-gfm and rehype-sanitize, so notes display beautifully and safely.
• Data layer: React Query and react-hook-form with zod validation.
• Design: a quiet-luxury look in deep forest green, ivory and champagne gold, with a serif reading face so a session note feels like a letter, not a database row.
• Test data: we wrote a realistic three-visit story (a patient with newly diagnosed type 2 diabetes) and generated it as two-speaker audio with free text-to-speech, so we could test the whole record, upload and note flow with consistent, fictional data.

Challenges we ran into

• Realistic test audio. Real consultations are private, so we had to script and voice our own, with different emotions, pauses and speakers, to make testing meaningful.
• Recording in the browser. Microphone permissions, differing audio formats across browsers (especially iOS Safari), and long files all needed careful fallbacks.
• Accuracy matters. A wrong dose in a note is worse than no note. We designed the product around the doctor as the final reviewer of every note.
• Privacy and trust. Patient records are sensitive. We designed consent reminders into recording and flagged that patient access by ID alone is weak, so a second factor (date of birth or a PIN) is a priority.
• Calm design for a stressful setting. We wanted something that felt like a private clinic, not enterprise software.

Accomplishments that we're proud of

• We built a complete flow for both sides of the consultation: a doctor records or uploads, and a patient later reads a clean note in the same place.
• We designed a calm, premium interface that doesn't feel like hospital software, with a reading experience built to feel like a personal letter.
• We created a realistic three-visit patient story with voiced, two-speaker audio, so we could test continuity across sessions without using any real patient data.
• We built privacy and safety into the design from the start, including consent reminders, sanitized note rendering and doctor review as the final check.
• We turned a simple observation, that important things said in a consultation get forgotten, into something a doctor can use in seconds before a follow-up.

What we learned

• The hardest part of care is often not diagnosis but continuity.
• A good medical summary is mostly about structure: what changed, what was decided, and what to watch for.
• Design is part of trust. Slow, quiet, readable interfaces make sensitive information feel safe.
• Realistic fake data is a powerful way to test health software without touching real patients.

What's next for Care Memory

• Build a dedicated physical recording device for the consultation room, so conversations are captured at professional audio quality without relying on a phone or laptop microphone.
• Highlight changes between sessions (for example, medication or weight changes over time).
• Add doctor review and edit before a note is shared with the patient.
• Add a second factor for patient access, plus audit logs.
• Support multiple languages, so consultations can be captured and understood in the language people actually speak.
• Add reminders so follow-ups and tests are not forgotten.

Care Memory: every conversation, remembered.

Built With

• fastapi
• gemini
• python
• react.js
• speech-to-text
• typescript
• whisper-ai