Medigo

Inspiration The idea for Medigo was born from a real need in West African communities: the lack of access to digital healthcare solutions for both urban and rural populations. In Senegal and neighboring countries, millions struggle daily to find available medications, schedule doctor visits, or adhere to treatment regimens—especially those without smartphones or reliable internet access.

We were inspired by the possibility of combining AI and USSD technology to bridge this digital divide and offer a healthcare platform that is inclusive, intelligent, and impactful. Our vision was to empower people, no matter their device or connectivity level, to manage their health more effectively.

What we learned Building Medigo was an intensive journey filled with technical and human lessons:

We learned to build OCR models tailored for French medical handwriting, which posed unique challenges in vocabulary, abbreviations, and formatting.

We gained experience in integrating geospatial queries for real-time pharmacy availability using PostGIS.

We discovered how to make USSD interfaces user-friendly and robust despite their limited UI capabilities.

And we deepened our understanding of medical data security, both technically and ethically.

How we built it Medigo is built on a modular architecture:

We started with user research and design sprints to understand healthcare pain points in West Africa.

The backend was developed using Django REST Framework with secure API endpoints and Celery for asynchronous task handling.

The frontend is a Flutter app, ensuring cross-platform support (Android/iOS), with a fallback USSD interface powered by a local gateway (*144#).

The prescription scanning feature uses a multi-stage AI pipeline:

OpenCV for image preprocessing

TensorFlow and Gemini 1.5 Flash for OCR and handwriting recognition

Custom-trained models to extract key data fields: ( \text{medicine_name}, \text{dosage}, \text{frequency}, \text{duration} )

We hosted our services locally with plans for cloud migration (AWS or GCP) as the platform scales.

Challenges Some of the biggest challenges we faced included:

Achieving high OCR accuracy with noisy and handwritten French prescriptions. We had to train and validate with real-world medical data and fine-tune our models to reach over 92% accuracy.

Integrating live pharmacy data in a region with limited APIs. We worked closely with pharmacies to manually update stock data, later automating through lightweight APIs.

Designing a powerful yet intuitive USSD experience, constrained by 160-character responses and no rich UI.

Ensuring data protection and regulatory compliance, especially for storing sensitive medical records and enabling encrypted doctor-patient messaging.

Despite these hurdles, we managed to deploy a working prototype that's already being tested by local healthcare partners.