🌱 About the Project

🔥 Inspiration

It started with Elizabeth, a 20-year-old girl from Lagos, Nigeria.
She developed a skin condition that her family believed was a curse.
By the time she saw a doctor, after months of shame and delay, the damage was permanent.

We realised Elizabeth isn’t alone.
Millions of people across Africa and other underserved regions face invisible medical barriers:
distance from doctors, misinformation, cultural stigma, language gaps, and sheer inaccessibility.

We asked:
What if health guidance could speak to people — literally — in the language they understand, right from their phones?

That’s how DermoAI was born — a mobile-first, AI-powered triage assistant that helps users:

Speak or type in their local language to describe symptoms
Upload or snap a photo of a skin condition for smart diagnosis
Understand their condition deeply — causes, triggers, treatments, risks
Get recommendations for care
Connect with the nearest verified clinics
Save and track their health history

🛠️ How We Built It

Frontend: Flutter, designed with accessibility and UX sensitivity in mind
Backend: Node.js/Express.js with MongoDB as the database and Cloudinary for secure image hosting

AI & Intelligence Layer:

Dermatological Analysis: A specialised deep learning model from Hugging Face (Autoderm), trained on diverse skin tones; Gemini 2.0 Flash serves as a fallback
Language Translation: Self-hosted LibreTranslate via Docker for real-time, private, multilingual support
Voice Transcription: Flutter SDK for converting voice-to-text, even in low-resource languages
LLM Guidance: Gemini Flash 2.0 for contextual analysis, health education, and symptom triage
Mapping & Navigation: OpenStreetMap integration for local clinic recommendations without premium APIs

💡 What We Learned

Trust is essential in healthcare — users need clarity, not just information
UX in healthcare is everything — We iterated heavily to make it intuitive for low-literacy users
Language and culture are health factors, not just barriers
You can build a FAANG-level experience with empathy and local insight — even with limited resources

⚔️ Challenges We Faced

Medical Safety Concerns: AI is not a doctor, so we built in explainable responses, disclaimers, and optional human review
Skin Tone Representation: Many datasets are biased — we’re incorporating diverse samples and exploring synthetic augmentation
Latency in Rural Areas: We optimised for low bandwidth using image compression and translation caching
Translation Quality: LibreTranslate was self-hosted and refined to ensure context-sensitive, culturally respectful output
Healthcare Alignment: DermoAI supports, not replaces, medical professionals — it provides triage and guidance, not final diagnosis