FrudgeCare AI
An end-to-end AI care-coordination platform built during the CareDevi AI Healthcare Innovation Hackathon 2026 (AI Patient Triage track). One free-text intake fans out into role-specific plans for the front desk, nurse, and provider — same input, three structured outputs.
Role
Team Lead
Stack
7 tools
Status
Prototype
Year
2026
Overview
An end-to-end AI care-coordination platform built during the CareDevi AI Healthcare Innovation Hackathon 2026 (AI Patient Triage track). One free-text intake fans out into role-specific plans for the front desk, nurse, and provider — same input, three structured outputs. Designed to fit into existing workflows, this build focuses on real production behavior rather than a polished demo.
The Problem
In a clinic, the same patient symptoms get re-asked at the front desk, by the nurse, and again by the provider. A prototype was needed to explore whether one structured AI intake could prepare role-specific outputs without becoming a black box for clinicians.
What I Built
- 01Single-input triage that generates simultaneous care plans for 3 staff roles
- 024-tier AI architecture with transparent source labeling
- 03FHIR R4-compliant care plan output
- 04Operations dashboard with real-time charts
- 05Full command palette for natural-language navigation
Architecture
→ Next.js + React + TypeScript client with MUI + Tailwind UI layer → FastAPI service for NLP, schema validation, and orchestration → Gemini 2.5 Flash Lite for structured symptom analysis → Regex + heuristic clinical NLP layer as deterministic fallback → ICD-10 + FHIR R4 references for clinically familiar outputs → Synthea-generated synthetic data — no real patient data used
Engineering Decisions
Strict structured outputs
Designed the AI layer to emit strict JSON schemas so clinical UIs can render deterministically instead of parsing freeform text.
Deterministic fallback path
Built around the assumption that AI may fail — a regex + heuristic clinical NLP layer keeps the system safe and reviewable when the model is unavailable.
Source-tier transparency
Each suggestion shows which tier of logic produced it (AI, heuristic, fallback) so clinicians can trust and audit the system.
Results & Learnings
- Working prototype demoed with clinician feedback in mind
- Structured JSON contracts make clinical UIs deterministic
- Fallback layer keeps the system useful when AI is unavailable
- Synthetic-only — explicitly not a medical device