MediBot – The AI-Powered Medical Assistant

• 

• 

• 

• 

• 

Inspiration

Every day, millions of people turn to the internet for health advice — often ending up misinformed, anxious, or worse. I wanted to build something better. Something human.

MediBot was inspired by one bold idea:

“What if getting medical help felt as real and comforting as talking to an actual doctor — instantly?”

I envisioned a tool that could speak, look, and feel like a doctor — providing expert advice in a friendly, conversational video format powered by AI.

I’m a solo founder and software developer from Nigeria who believes in building impactful tools for Africa.

MediBot was inspired by the reality that most doctors in underserved regions have to manually interpret scattered patient data. I wanted to fix that by combining AI and automation to support real-time medical insights—even in low-resource settings.

I had never worked with large language models locally before, but I learned how to use Ollama, set up custom endpoints, and generate health-focused summaries with the Meditron model. I even integrated Tavus to turn data into videos and used ElevenLabs for accessibility.

This project represents not just a hackathon submission—but a future vision of AI improving African healthcare. Building MediBot under time pressure, learning on the go, and seeing it work locally was a major moment of growth.

My hope is that this project inspires other developers across the continent to build AI-first solutions for real local problems.

What it does

MediBot allows users to input health concerns in natural language — like “I’ve had chest pain and shortness of breath.” Behind the scenes, It analyze that input, generate a personalized medical response, and deliver it as a realistic AI doctor video with voice using:

🎥 Tavus – for lifelike AI-generated doctor video.

🗣️ ElevenLabs – for ultra-human voice synthesis.

🧠 Ollama (MedLM/Med-Gemini) – for medical-grade AI reasoning.

Users can follow up with more questions, just like a real consultation. MediBot responds again, in seconds — with video, voice, and medical logic.

How we built it

I built MediBot from the ground up using:

React + Vite for a snappy, responsive frontend.

Node.js + Express for a lightweight backend.

Ollama (local LLM) with the meditron model to process medical input.

Tavus API to generate video of a lifelike doctor speaking the AI’s response.

ElevenLabs API to convert the AI’s response into human-sounding audio.

I also implemented real-time polling to update video generation status and ensure seamless UX during async AI video creation.

Challenges We Faced

Tavus API limitations made us redesign our async workflow.

Voice-video sync was tricky — I had to align ElevenLabs’ audio output with Tavus’ video response timeline.

Data privacy concerns pushed us to keep all processing local with Ollama instead of calling OpenAI or cloud-hosted LLMs.

I had to balance realism and response time — fast enough to feel helpful, but high quality enough to feel real.

Polling logic: Ensuring the frontend checks Tavus job completion and plays the video smoothly

Accomplishments that we're proud of

✅ Built a full AI-powered doctor with talking head and human voice in 48 hours

✅ Delivered personalized medical advice via video in under a few minute

✅ Created a scalable foundation for future integrations like medical records, triage routing, and more

✅ Pushed the limits of what “chat with a doctor” could feel like

What we learned

Tavus makes it surprisingly easy to turn text into believable, expressive video

ElevenLabs voice cloning feels incredibly natural — a game-changer for emotional AI

Seamless AI experiences depend on more than models — UX flow, timing, and polish matter just as much

The future of healthcare isn’t just automation — it’s humanization through AI

What's next for MediBot – The AI-Powered Medical Assistant

🏥 Integrate real EHR data (electronic health records) for doctor-specific analysis

🌍 Support more languages and accents via ElevenLabs and Tavus

👨‍⚕️ Add human-in-the-loop review for critical symptoms

📱 Launch on mobile as a 24/7 health companion for underserved regions

Built With

• bolt.new
• elevenlabs
• express.js
• meditron)
• node.js
• ollama
• react
• tavus