SymptoSense

Logo

Inspiration

Doctors are drowning in paperwork, and patients struggle to remember medical details during short appointments. Most health data stays "trapped" in messy voice notes or mental fatigue. We built SymptoSense to ask: What if patients could just talk, and AI handled the rest?

What it does

It’s a dual-interface platform that turns natural speech into structured medicine.

For Patients: A private diary to log symptoms, meds, and mood via voice. AI extracts clinical entities and analyzes vocal tone for stress or fatigue.

For Doctors: Converts patient logs into structured SOAP notes with AI-suggested diagnoses and plans. A severity dashboard prioritizes critical cases, and SHA-256 hashing ensures prescriptions are tamper-proof.

How we built it

We used Next.js 14 and Supabase for a secure, role-based backend. The "brain" is a combination of Azure Speech SDK for medical-optimized transcription, Azure Text Analytics for Health for extracting symptoms/dosages, and GPT-4 to synthesize everything into clinical SOAP notes.

Challenges we ran into

Medical Accuracy: Raw speech-to-text often mangles drug names. We had to layer health-specific NLP on top to ensure "Lisinopril" didn't become a random word.

Data Isolation: Using Supabase required us to build a custom role-based API layer to ensure patient diaries stay 100% private from doctors unless explicitly shared.

The "AI Override": We had to architect a "Reject & Rewrite" flow so doctors could instantly wipe AI suggestions if they disagreed, keeping the human in total control.

Accomplishments that we're proud of

Clinical Safety: Our "Reject & Rewrite" isn't a gimmick; it’s a legitimate safety feature that protects physician judgment.

Vocal Biomarkers: We successfully implemented tone analysis to detect patient stress and fatigue, adding a layer of data text can't capture.

Tamper-Proofing: Using SHA-256 hashing for digital receipts provides blockchain-level security without the overhead.

What we learned

AI in healthcare must be assistive, not authoritative. The moment we gave doctors the power to override the AI, the tool became more trustworthy. We also learned that healthcare-specific AI models (like Azure’s) are vastly superior to general-purpose ones when handling complex dosages and symptoms.

What's next for SymptoSense

We are working on FHIR R4 Integration to sync data with existing hospital EHR systems. We’re also building a Symptom Correlation Engine to help patients spot patterns, like how lifestyle factors or missed meds trigger specific flare-ups.