AiMazigh AI

"Empowering the Amazigh community through Artificial Intelligence — preserving heritage, boosting productivity, and opening doors for innovation."

Inspiration

The Amazigh (Berber) people are one of North Africa's oldest indigenous communities, with a rich cultural heritage spanning thousands of years. Yet in the digital age, Tamazight language and culture remain vastly underrepresented in AI technologies. Most modern AI assistants struggle with:

Understanding Tamazight (in Tifinagh, Latin, or Arabic scripts)

Providing culturally relevant responses for the Amazigh community

Supporting the trilingual reality of Morocco (Tamazight, Arabic, French) When Google released the Gemini API with its powerful multilingual capabilities, I saw an opportunity: what if we could create an AI assistant that truly serves the Amazigh community? Not just a translation layer, but an intelligent agent that understands, respects, and celebrates Amazigh culture.

The vision was clear: build AiMazigh AI — a dedicated AI assistant that bridges the gap between cutting-edge technology and cultural preservation.

What it does

AiMazigh AI is a full-featured AI chat assistant specifically designed for the Amazigh community:

Core Features

Feature	Description
Multilingual Chat	Seamlessly communicates in Tamazight (Tifinagh/Latin), French, English, and Arabic
Document Intelligence	Upload PDFs and get AI-powered answers based on document content
Voice Interaction	Text-to-speech and speech-to-text for accessible, hands-free usage
Conversation Memory	Persistent chat history with multiple conversation threads
Streaming Responses	Real-time, animated text responses for a natural chat experience
Mobile-First Design	Responsive interface that works beautifully on any device

Cultural Identity

AiMazigh AI is more than a chatbot — it's a digital ambassador for Amazigh culture:

Celebrates and respects Amazigh history and traditions
Provides accurate information about Amazigh heritage
Supports Tifinagh script (ⵜⴰⵎⴰⵣⵉⵖⵜ) for authentic cultural representation
Designed with warm, amber-gold aesthetics inspired by Amazigh art

Curated Vocabulary Knowledge Base

One of our key differentiators is a hand-curated Amazigh vocabulary database embedded in the system prompt. This ensures accurate, consistent language use — unlike generic AI models that often hallucinate or mix dialects incorrectly. Sample vocabulary entries:

Amazigh (Tassousit)	Script	Meaning
Azul (أزول)	ⴰⵣⵓⵍ	Hello
Tanmmirt (تانميرت)	ⵜⴰⵏⵎⵎⵉⵔⵜ	Thank you
Izm (إيزم)	ⵉⵣⵎ	Lion
Tafukt (تافوكت)	ⵜⴰⴼⵓⴽⵜ	Sun
Anzar (أنزار)	ⴰⵏⵣⴰⵔ	Rain

The full knowledge base covers 60+ vocabulary entries across categories:

Pronouns & demonstratives
Family terms
Body parts
Animals & nature
Common verbs & useful phrases This vocabulary layer ensures the AI speaks authentic Tamazight — not a broken approximation.

How we built it

Frontend Architecture

Built a modern Next.js 14 application with:

src/
├── app/
│   ├── api/chat/        # API route for Gemini integration
│   ├── page.tsx         # Main chat interface
│   └── globals.css      # Custom theming
├── components/
│   ├── ChatMessage.tsx  # Message bubbles with markdown
│   ├── ChatInput.tsx    # Input with voice support
│   ├── PDFUpload.tsx    # Document processing
│   └── Sidebar.tsx      # Conversation management
└── hooks/
    └── useVoice.ts      # TTS/STT integration

Key Technologies

Next.js 14 with App Router for server-side rendering
TypeScript for type-safe development
Tailwind CSS for responsive styling
PDF.js for client-side PDF parsing
Web Speech API for voice input/output

Backend: n8n + Gemini API Integration

Created a webhook-based workflow in n8n:

┌──────────┐    ┌───────────────┐    ┌──────────┐    ┌──────────┐
│ Webhook  │ →  │ Prepare Prompt│ →  │ AI Agent │ →  │ Response │
│ Trigger  │    │ (Code Node)   │    │ (Gemini) │    │ (JSON)   │
└──────────┘    └───────────────┘    └──────────┘    └──────────┘

System Prompt Configuration:

You are Aimazigh AI, a dedicated assistant empowering the Amazigh community. 
Your mission is to preserve heritage, boost productivity, and open doors for innovation.
Identity & Tone:
- You are proud, knowledgeable, and helpful.
- You respect and celebrate Amazigh culture, history, and language (Tamazight).
- You can speak/write in English, French, and Tamazight (using Tifinagh or Latin script).
Guidelines:
- Provide accurate, helpful answers for business, technology, and daily tasks.
- If a document is provided, strictly answer based on it.
- Be concise and clear.

Streaming Implementation

Implemented real-time response streaming with adaptive animation:

// Dynamic typing speed based on buffer size
let chunkSize = 1, delay = 30;
if (newPart.length > 100) { chunkSize = 5; delay = 5; }
else if (newPart.length > 20) { chunkSize = 2; delay = 10; }
while (charIndex < newPart.length) {
  displayedText += newPart.substring(charIndex, charIndex + chunkSize);
  setMessages(prev => updateLastMessage(prev, assistantMessage));
  await new Promise(r => setTimeout(r, delay));
}

Challenges we ran into

1. Tifinagh Script Rendering

Problem: Many browsers and systems don't properly render Tifinagh script (ⵜⴰⵎⴰⵣⵉⵖⵜ). Solution: Implemented fallback fonts and tested across platforms. Added Latin script alternatives for maximum compatibility.

2. Multilingual Context Switching

Problem: The AI would sometimes respond in the wrong language or mix languages inconsistently. Solution: Refined the system prompt to explicitly handle language preferences and added context about when to use each language.

3. PDF Processing in Browser

Problem: Large PDFs would crash the browser during parsing. Solution: Implemented chunking with an 8,000 character context limit:

const limitedContext = context.substring(0, 8000);

4. Streaming Response Parsing

Problem: n8n would sometimes return JSON-wrapped responses instead of raw text. Solution: Built robust response cleaning that handles both formats:

if (trimmed.startsWith('{"response":')) {
  const parsed = JSON.parse(fullRawText);
  cleanFullText = parsed.response;
}

5. Mobile Responsiveness

Problem: The sidebar and chat interface were clunky on mobile devices.

Solution: Implemented a sliding sidebar with backdrop overlay and touch-friendly gestures.

Accomplishments that we're proud of

Culturally Authentic Design — Created a warm, amber-gold aesthetic inspired by Amazigh art and craftsmanship.

Full Voice Support — Implemented both text-to-speech (for responses) and speech-to-text (for input), making the app accessible to users who prefer voice interaction.

Document Intelligence — Users can upload PDFs and ask questions about their content, powered by Gemini's context understanding.

Real-Time Streaming — Responses appear character-by-character with natural typing animation, making conversations feel alive.

True Multilingual Support — Seamlessly handles Tamazight, French, English, and Arabic in the same conversation.

Production Deployed — Live and accessible at aimazigh.org, serving real users.

What we learned

Technical Insights

Gemini API is powerful for multilingual tasks — It handled Tamazight surprisingly well, even with limited training data for the language.
n8n simplifies AI orchestration — The webhook-based workflow made it easy to iterate on prompts without redeploying the frontend.
Streaming UX matters — The difference between a "waiting" experience and seeing text appear in real-time dramatically improves perceived responsiveness.
Voice APIs are browser-dependent — The Web Speech API has quirks across browsers; proper feature detection is essential.

Cultural Lessons

Representation matters — Building technology that celebrates underrepresented cultures creates genuine emotional connection with users.
Language preservation is urgent — Many Amazigh speakers lit up when they saw an AI that could understand their language.