Medora
AI for your health.
A multimodal health AI platform that unifies image, audio, signal, and text processing under one roof. Designed and built end-to-end on FastAPI, React, Flutter, and PostgreSQL — running today on Docker.
Medora in numbers
Problem & Solution
Health data is not one thing: a prescription photo, a skin lesion image, a heart-rate series, and a patient's complaint in free text all belong to the same person, yet each demands a completely different processing technique. Most systems pick one and leave the patient with a fragmented experience.
Medora unifies these modalities in a single platform: computer vision reads the prescription and the lesion, Whisper transcribes speech, a signal-processing layer catches anomalies in vital measurements, and a health-focused Turkish voice agent translates all of it into language the patient understands — with 112 emergency routing and a strict “not medical advice” safety frame. Patient, doctor, clinic manager, and admin each get their own role-based interface, and web and mobile deliver a pixel-identical experience.
Competency Matrix — what exactly was built in Medora?
This project is evidence of end-to-end engineering breadth, not a single specialty.
01Computer Vision / Deep Learning
- Skin lesion analysis: lesion type, malignancy risk, class scores, and a see-a-doctor recommendation (gpt-4o vision)
- X-ray image analysis
- Prescription OCR: extracting drug name + dose + usage from real prescription images at 98–99% confidence, mapped to a drug database and pushed into e-prescriptions.
02Audio Processing
- Speech recognition with Whisper
- spoken responses via tts-1 with auto-playback on every turn (voice-first UI)
- real-time audio streaming over WebSocket
- microphone-permission fallbacks.
03Signal Processing
- Heart rate, blood pressure (systolic/diastolic merging), temperature, SpO₂, and glucose streams
- anomaly detection and automatic alert generation over 3,300+ measurements (vital_anomaly service).
04Machine Learning
- 10 disease prediction models (GradientBoosting / sklearn pipelines)
- the heart model trained on real data
- a dedicated Flask ML serving layer.
05LLM & Agentic AI
- A health-focused Turkish voice agent (gpt-4o-mini)
- 112 emergency routing and a “not medical advice” safety disclaimer
- RAG with semantic search on pgvector
- agentic report generation
- a configurable LLM layer (local/remote models + A/B testing).
06Backend & API
- 201 routes on FastAPI
- async SQLAlchemy
- real-time WebSocket channels
- rate limiting
- nginx reverse proxy.
07Database
- PostgreSQL + pgvector
- Redis
- a 17-module idempotent demo seed system
- a realistic dataset of 22 active patients, 8 doctors, and thousands of clinical records.
08Web & Mobile
- React/Vite web (~85 kB main bundle, code-split)
- Flutter mobile: 30+ screens, pixel-mirrored against the web app
- flutter analyze: 0 errors / 0 warnings
- 14 custom web UI components + 10 Medora Flutter widgets.
09Security & Identity
- JWT + argon2/bcrypt
- MFA
- role-based access (patient / doctor / clinic manager / admin)
- WebSocket auth (Sec-WebSocket-Protocol: bearer.<jwt>)
- rate limiting.
10Healthcare Integrations
- FHIR
- e-Nabız
- e-Prescription
- e-Report
- MHRS
- an e-Devlet flow
- billing + e-Invoice
- PDF generation with full Turkish character support (embedded DejaVu fonts)
- Google Meet video consultations
- family profiles
- medication reminders with adherence scoring
- a moderated community module.
11Accessibility & Design System
- The “Modern Medical” design system (Medical Teal #0EA5A6, Inter)
- WCAG 2.1 AA: modal focus traps + ESC, arrow-key tab navigation, skip-to-content link, aria-labels
- measured contrast fixes (6.09:1)
- page transitions and KPI stagger animations.
Architecture
Medora is a containerized system of five cooperating services: the FastAPI backend (201 routes), the React/Vite web client, a Flask ML server, PostgreSQL (with pgvector), and Redis — fronted by nginx. The Flutter mobile app consumes the same API. Three separate Docker Compose profiles (local / dev / prod) and multi-stage builds carry the same codebase from development to production; all five containers run healthy.
Docker Compose ×3 profiles (local / dev / prod) · multi-stage build · 5/5 containers healthy
Gallery
Screenshots are taken from the live running system; patient data is demo seed data.
Technology Stack
AI/ML
Backend
Frontend & Mobile
Data & Infrastructure
Quality
Engineering Quality
- Playwright E2E tests; a clean baseline across 4 roles × ~60 routes.
- A stable release after multiple bug-hunt and fix cycles.
- flutter analyze: 0 errors / 0 warnings.
- WCAG 2.1 AA verified by measurement (including 6.09:1 contrast fixes).
- A 17-module idempotent seed: the system boots with realistic data in any environment with a single command.
- 3 environment profiles (local/dev/prod), multi-stage Docker builds, 5/5 containers healthy.
Confidentiality Note
Medora's source code is closed for commercial and confidentiality reasons. A live demo, an architecture walkthrough, and a code review can be arranged upon request.
Get in touch for detailed projects like this one
Medora is just one of 15+ projects under NDA. Let's talk about what I can bring to your team.