LUCA Nursing Assessment App
A full-stack Progressive Web App for novice nurses featuring AI-powered clinical recommendations via a custom weighted KNN retrieval model.
Project Overview
This application was built for a newly founded business aimed at solving the problem of novice nurses feeling overwhelmed and underprepared for patient assessments. My senior design team was tasked with building a minimum viable product to be released to a pilot group of nurses for feedback and validation.
As the development lead, I guided the team in designing and building the full-stack webapp, defining technical requirements and architecture. What resulted is a user-friendly application that guides new nurses through a structured patient assessment flow, collecting vitals and symptoms to generate clinical recommendations in real time.
Recommendations are produced by a custom weighted KNN retrieval model trained on verified clinical cases. Patient data is encoded into a feature vector with clinically prioritized weights, ensuring outputs are grounded in validated cases rather than generative inference — achieving 100% accurate outputs with no hallucination risk.
Technical Features
- •Guided Assessment Flow: Multi-step patient intake form collecting age, signs & symptoms, and vitals to build a structured feature vector for accurate model input
- •Account System: Full authentication flow with secure registration, bcrypt password hashing, JWT session management, and email-based password reset via SendGrid
- •React Frontend: Responsive Progressive Web App built with Vite and Tailwind CSS, installable on any device without an app store
- •FastAPI Backend: Python REST API handling model inference, session validation, and account management with Pydantic-enforced request validation
- •Supabase Integration: Postgres database via Supabase storing nurse accounts and sessions, with token expiry enforced at the query level for secure, server-side session management
App Walkthrough
The entry point routes users to login or registration. Signup collects name, email, phone, date of birth, and specialty unit with inline validation and real-time password requirements.
The assessment begins by bucketing the patient into an age range, which feeds into the model. Numeric vitals are then entered — heart rate, blood pressure, temperature, respiratory rate, and SpO2 — each validated before proceeding.
Signs and symptoms are collected across five clinical categories — color, respiratory, abdomen, alarms, and neuro — using single and multi-select modals, with an optional lab values field. The summary screen renders the full collected payload for review before submitting to the KNN model.
