Inspiration

Our elders already live on WhatsApp—it's how they talk to family and community every day. But basic health info (which clinic, when it’s open, who to call) is scattered or unclear, especially outside big cities. We wanted a simple, friendly chat experience that helps elders find care without installing new apps.

What it does

AfyaBot is a WhatsApp chatbot that understands messages like “clinic in Mekelle” and replies with 2–3 nearby clinic suggestions from a small curated directory (JSON).

It supports English ↔ Amharic/Tigrinya switching and keeps replies short, clear, and elder-friendly (ending with one simple question).

How we built it

Channel: WhatsApp via Twilio webhook. Backend: Node.js + Express. AI: OpenAI API (LLM handles intent, language, tone; we also added a safety-net for clinic queries). Data: Single embedded JSON file shipped with the service (no database). Dev: Local testing with curl/ngrok; clean service modules for AI + directory search.

Challenges we ran into

Grounding the LLM so it never “makes up” places → we only return entries that exist in our JSON directory. Coverage gaps for smaller towns → built a simple scoring/search on city/district/neighborhood and a graceful “no data yet” response. Language UX for elders → tuned prompts for short messages, simple words, and one follow-up question. WhatsApp payload quirks (JSON vs form fields) → unified parsing for Body, Latitude, Longitude.

Accomplishments that we're proud of

End-to-end WhatsApp flow working with clean, empathetic replies. Reliable clinic suggestions (2–3 results) from a lightweight, maintainable JSON directory. Language switching that feels natural, not “techy.” A small codebase that’s easy to read, extend, and demo.

What we learned

Meet users where they already are: WhatsApp > new apps for elders. Short, clear language matters more than fancy UI. Data curation is a product feature—quality of the directory drives user trust. Prompt design + guardrails are essential when using LLMs for real users.

What's next for AfyaBot

Grow the JSON clinic directory (community/clinic contributions + simple admin review). Add per-chat preferences (language, favorite clinics) and basic usage metrics. Ship Twilio XML responses for production WhatsApp delivery and deploy a stable webhook. Partner pilots with local clinics/community orgs to validate utility and fill data gaps.

Share this project:

Updates