Pneuma

Pneuma: Vitals in a snap.

Comment

Pneuma: Vitals in a snap (aka TriageAid v2)

Github

https://github.com/Darksharkthe1st/Pneuma.git

Track:

Health and Wellbeing

Who you built this for and why they need it:

We built Penuma for community health workers and nurses operating in underserved or remote clinics. In low-resource healthcare settings, basic triage—taking vitals and recording initial symptoms—creates massive bottlenecks because it requires dedicated equipment and staff time. Patients often wait hours just to be seen for an initial assessment. They need a rapid, accessible, and equipment-free way to collect baseline patient data so that doctors can prioritize care based on immediate medical need, rather than who arrived first.

How you used Claude / AI in your project:

We used the Claude API specifically for symptom summarization and data structuring. When a patient inputs their symptoms in plain, conversational language (e.g., "My stomach has been hurting really bad since yesterday and I feel super hot"), we pass this text to Claude. Claude then translates this raw input into a concise, professional medical summary card (e.g., "Patient reports severe abdominal pain and fever onset 24 hours ago"). This allows the doctor to instantly digest the patient's condition without reading paragraphs of unorganized text.

What could go wrong and how you addressed it

AI Hallucinations in Triage: The biggest risk is an AI model incorrectly diagnosing a patient or assigning them the wrong triage urgency, which is a massive medical liability. How we addressed it: We explicitly walled off the AI from making any medical decisions. We built a deterministic, hard-coded rules engine in JavaScript that assigns the urgency color (Red/Yellow/Green) based strictly on heart rate thresholds and specific keywords. Claude is strictly limited to formatting text.

Data Privacy (HIPAA): Taking photos of patients raises severe privacy concerns. How we addressed it: Our prototype is designed so that the camera feed used for the Presage heart rate calculation is processed ephemerally. No images of the patient are stored in a database or sent to any external API. Hardware Biases: Remote photoplethysmography (rPPG) camera tech can sometimes struggle in low lighting or with darker skin tones. How we addressed it: We designed the system to allow health workers to manually input or override the heart rate if the camera fails to get a confident reading.

What you'd build next if you had more time:

Multilingual & Audio Support: We would add voice-to-text functionality so patients with low literacy or different language backgrounds could simply speak their symptoms into the tablet, which Claude would translate and summarize in the doctor's native language.

Expanded Vitals (Respiratory Rate): Since rPPG technology can also track breathing patterns, we would expand the camera integration to calculate respiratory rate alongside heart rate for a more complete triage profile.

EMR Integration: We would build webhooks to export the final triage summary directly into open-source Electronic Medical Record (EMR) systems like OpenMRS, which are widely used in developing nations.

Built With

Share this project:

Updates