MediLexi

Inspiration

In many parts of Africa, access to qualified healthcare professionals, diagnostic tools, and reliable drug information is severely limited—especially in rural or underserved regions. Clinics and health posts often lack internet access, medical libraries, or support staff, leaving health workers to make difficult decisions based on limited information. We were inspired to build MediLexi—an offline, AI-powered medical assistant—to bridge this healthcare information gap in a low-power, resource-constrained environment.

What it does

MediLexi is an offline, Python-based AI assistant that: Diagnoses symptoms through natural language input and AI classification Provides drug information, dosage guidelines, and interaction warnings Speaks out medical instructions using offline text-to-speech in multiple local languages Runs entirely offline after setup, even on low-end devices, making it perfect for use in rural clinics, mobile outreach vans, or during network blackouts

How we built it

MediLexi is built using Python and structured as a lightweight desktop application: We used DistilBERT, a small transformer model, to classify symptom descriptions into likely conditions. SQLite powers the offline medical drug database, with thousands of Africa-relevant drug entries. We used pyttsx3 for completely offline voice playback of instructions, which is crucial in low-literacy environments. The interface is built using Tkinter, chosen for its portability and simplicity. The app is bundled as a standalone executable with PyInstaller, requiring no installation or internet connection after setup.

Challenges we ran into

Offline AI limitations: Deploying AI locally without cloud resources required extensive model compression and performance tuning. Medical accuracy: Designing a simple but accurate diagnostic logic system that balances safety and usability was critical. Voice synthesis: Ensuring offline TTS worked across devices, accents, and languages with clarity was challenging. Data sourcing: Gathering curated, localized medical data (especially for drug usage and dosage relevant to African treatment protocols) required significant validation.

Accomplishments that we're proud of

Created a fully functional offline AI assistant for healthcare—completely runnable without cloud or external APIs Ensured the solution works on low-spec devices, including recycled laptops and entry-level Android tablets Integrated offline multilingual voice playback, improving accessibility for non-literate or visually impaired users Packaged the solution into a portable application requiring no tech expertise to use

What we learned

How to compress and optimize AI models for use in extreme compute-constrained environments The power of combining local storage, AI, and UI simplicity for real-world deployment Deep insights into the challenges faced by frontline health workers in low-infrastructure settings How offline-first design thinking can lead to resilient, scalable solutions in areas where traditional apps fail

What's next for MediLexi

Multilingual support: Integrate additional voice packs and interfaces for Swahili, Hausa, Yoruba, and French Mobile version: Deploy a PWA or Kivy-based Android app version for phone-based community health workers Local update syncing: Enable USB/SD card-based update delivery for clinics with no internet NGO & public health collaboration: Partner with healthcare NGOs and ministries to preload MediLexi on public health devices Symptom escalation logic: Add urgent-condition flagging that recommends immediate transfer or teleconsultation

Built With

• python