ElderCare Companion Summary (Production-Ready SaaS)

Multi-tenant platform for Canadian personal support agencies and long-term care facilities. Built with FastAPI (Python 3.13) backend, React Native/Expo 54 frontend. Fully tested (537 backend, 379 frontend tests), 6 CI/CD workflows—all green. PIPEDA-compliant for 1.5M+ Canadian seniors’ care needs, replacing paper binders, WhatsApp, and spreadsheets to close compliance gaps and prevent incidents.

Why It Matters
Connects admins, nurses, PSWs, and families via single source of truth for patient care.

Core Features

Task Management: Schedule tasks; GPS-verified completion (lat/lon logged); clinical/non-clinical gating.

Health Monitoring: Log vitals/mood/meals; trend charts (BP, SpO₂, glucose) with badges (Rising/Falling/Stable) and normal ranges.

Medication Tracking: Digital MAR; role-gated admin; full audit logs.

Emergency Alerts: One-tap alerts; Redis idempotency for auto-escalation (tier-1 to tier-2).

Shift Management: GPS clock-in/out (100m radius); supervisor overrides with immutable audits.

Care Planning: Goal tracking (% progress, milestones, status).

Real-time Messaging: WebSockets (Live/Polling); HTTP fallback.

Photo Documentation: Wound/life photos (5MB, MIME-validated); family sharing toggle.

Incident Reporting: Severity levels (low/high/critical); photo attachments, follow-ups.

Smart Scheduling: AI-ranked staff by proximity/workload/consistency.

Push Notifications: Expo alerts for emergencies/shifts/tasks.

Family Portal: Read-only dashboard, snapshots, gallery, messaging.

Offline Support: Encrypted local queue with auto-sync.

i18n: English/French (200+ strings).

Onboarding: 5-step flow for Facility/Home Care.

Payroll: Timesheets from shifts; approvals, per-role rates, OT detection (ON 44h/wk), CSV for ADP/etc.

Data Export: PIPEDA patient record JSON exports.

MapleMatch is your AI-powered SaaS matching Canadians to affordable housing, subsidies, co-ops, and waitlists across all 13 provinces/territories. Live at web-eight-gamma-54.vercel.app (React frontend) with FastAPI backend at maplematch-api.onrender.com/docs. Pulls from 6 open-data CKAN sources + 74 curated seed listings for always-on fallback.

Core Value
Users input eligibility (income, household, location) via wizard; get instant AI-ranked matches by income fit, size, priority group, and predicted wait times—solving Canada’s housing crisis without paid APIs.

Key Features

Clerk JWT auth with auto-provisioning.

Eligibility wizard + real-time listing filters (province/city/rent/RGI/accessibility).

AI 3-layer engine: rules + semantic + ML scoring.

Accept/decline tracking + notification bell.

Proto-KYC: doc uploads + admin review.

Fully bilingual EN/FR; WCAG-accessible (ARIA, screen readers).

Admin dashboard: pending docs, CMHC sync history.

Rate limiting (120 req/min); daily GitHub Actions data sync.

Data Pipeline
Automated pulls from: Toronto CKAN (Affordable Rentals), CMHC Portal, Ontario/BC/Alberta/Montreal catalogues. Ensures comprehensive national coverage.

Flask-based web app predicts ASD likelihood in toddlers using Dr. Fadi Thabtah’s Q-Chat-10 dataset (1054 records, 18 features: A1-A10 behavioral scores + age/sex/ethnicity/jaundice/family history). Achieves 99.7% accuracy via Logistic Regression—deployable locally or Dockerized for healthcare pros/parents.

Tech Stack & Structure

Model Pipeline

One-hot encode categoricals (Sex/Ethnicity/Jaundice/Family_ASD)

StandardScaler normalization

LogisticRegression (5-fold CV: 99.7% accuracy)

Web form → predict → “Yes/No” ASD traits + probability

Key Features

Web Interface: Input A1-A10 scores + demographics → instant prediction

Retraining: python train.py regenerates model.pkl + ROC/confusion plots

Production-ready: Dockerized, virtualenv, requirements.txt

Interpretability: Logistic Regression coefficients explain feature impact

End-to-end anomaly detection system for high-frequency manufacturing sensor data (pressure/force/acceleration/temperature at 10ms sampling). Processes streams through signal filtering, feature extraction, and hybrid ML models (Isolation Forest baseline + Autoencoder/LSTM). Served via FastAPI /predict endpoint with PostgreSQL persistence and optional AWS SageMaker deployment.

Architecture Flow

Sensor Streams → Preprocessing (Butterworth LPF, optional Kalman, 50ms resampling)

Windowing (1s windows, 250ms stride)

Feature Extraction (RMS, kurtosis, spectral energy, ZCR, moving avg, peak-to-peak)

Anomaly Models ↔ PostgreSQL (Isolation Forest + AE/LSTM reconstruction)

FastAPI Service → MES/UI clients

Production Deployment

Local: pip install -e . then uvicorn psa.api.main:app --reload

Docker: docker compose up --build (API + Postgres)

API Payload: 10s sensor batches via POST /predict

Response: anomaly_score (0-1), is_anomaly (bool), per-window details

SageMaker-ready: Training (sagemaker/train.py), inference (sagemaker/serve.py), model artifacts in /opt/ml/model/iforest.joblib

Key Engineering

Real-time capable (50ms resampled windows)

Redis caching, rate limiting, immutable audit trails

Hybrid ML: Isolation Forest (fast baseline) + deep learning (temporal patterns)

Cloud-agnostic: Local/Docker or AWS S3→SageMaker pipeline

Overview
Your production-grade AI app classifies burn severity (1st/2nd/3rd degree) from images using TensorFlow deep learning + ensembles. FastAPI backend + Streamlit UI, PostgreSQL DB, full CI/CD via Azure DevOps, Dockerized on Docker Hub (emekamikolo777/SkinBurnPro). Supports auth, uploads, feedback, logging/reports.

Key Features

Burn Classification: Image upload → instant severity prediction

Auth: JWT

API Endpoints

User Flow: Upload → classify → feedback → downloadable report

Infra: SMTP alerts, env vars (DB/Gmail), comprehensive logging

Tech Stack

ML: TensorFlow (ensemble CNNs)

Backend: FastAPI + PostgreSQL

Frontend: Streamlit (interactive UI)

DevOps: Azure CI/CD, Docker multi-container (API/DB/Streamlit)