RAHA.ai

💡 Inspiration

In Morocco, many elderly people face loneliness, missed medications, and limited access to medical care ,often worsened by illiteracy or language barriers and heavily dependent on their children, often times ,occupied with their own family or work . Our own grandparents struggled with remembering appointments, understanding prescriptions, or even calling for help. We wanted to create something that truly cared ,not just a tool, but a companion.

That’s how Raha was born , from the Moroccan phrase "raha dyal l-bal" (راحة ديال البال), we aimed to provide peace of mind for their loved ones, as Raha will take care of their parents/grandparents even with them being busy.

🧠 What it does

Raha.ai is an AI-powered personal and medical assistant designed for Moroccan elders. It offers:

Medication & doctor appointment reminders
Voice interaction in Darija and Tamazight
Simplified explanations of medical terms when visiting the doctor or buying meds
Coordination of transport (taxi/inDrive)
Health tips, daily check-ins, and activity suggestions
A connected fall-detection & anti-lost bracelet (for dementia or alzheimer patients)
Emergency contact alerts for family or authorities

How we are building it

Frontend: We designed a voice-first interface using Flutter for mobile (elder-friendly, high contrast, simple layout) and React for caregiver web dashboards. All key actions can be triggered via voice to assist illiterate users.
Backend: Built on Node.js + Firebase, allowing real-time data sync for reminders, fall alerts, and emergency status. Firebase Cloud Functions trigger immediate alerts to family or authorities if the user is lost or falls.
Voice AI (Darija + Tamazight):
- Speech-to-Text: We fine-tuned OpenAI Whisper on Darija and Amazigh datasets (radio, YouTube, family recordings) to allow elders to speak naturally in their dialects.
- Text-to-Speech: Used Coqui.ai (Mozilla TTS fork) and local audio corpora to synthesize responses in Darija and Amazigh, making the assistant emotionally and linguistically accessible.
- For offline fallback, we used lightweight models (Vosk for STT and PicoTTS) that can run on low-end mobile devices.
- A custom translation layer maps Darija and Tamazight into Modern Standard Arabic, enabling GPT-4 to process requests, then translates the answer back into dialect using prompt-tuned phrasing.
AI Integration:
- Integrated GPT-4 via API for medical Q&A, appointment instructions, and friendly conversations.
- Built a custom intent classifier (TensorFlow Lite) for detecting commands like “medication,” “call doctor,” or “what’s my schedule?” from voice input.
Wearable Device:
- Developed an ESP32-based bracelet equipped with a gyroscope, GPS, and fall detection algorithm.
- If a fall or wandering is detected, it sends geolocation to the Firebase backend, which triggers alerts to caregivers.
Emergency System:
- Firebase Cloud Functions handle real-time emergency communication via SMS, push notifications, or automated voice calls.
Localization:
- We built an internal phrasebook engine with culturally adapted responses and medical explanations tailored to Moroccan elders.
- The assistant’s tone, expressions, and examples are localized to reflect Moroccan values and daily life.

Challenges we ran into

Speech recognition for dialects (Darija/Tamazight) was poorly supported, so we fine-tuned models and added custom phrase libraries.
Hardware sync between the wearable and the mobile app was tough under time pressure.
Designing for elders meant reducing interface complexity, increasing contrast, and emphasizing voice interaction.
Simplifying medical language while staying accurate took extensive prompt engineering.