Hagere Voice

Hagere Voice: Voice-first inventory management in Amharic(Ethiopia) empowering women artisans to manage stock offline.

Comment

Inspiration

We didn't start with a problem we read about in a report—we started with conversations around our dinner tables.

Our mothers, sisters, and neighbors are master tailors and artisans here in Ethiopia. They can craft a dress with precision, manage dozens of clients by memory, and keep their businesses afloat through sheer resilience. Yet despite their incredible skill and work ethic, they remain economically invisible. No digital footprint. No inventory system. No way to scale beyond their immediate community.

When disruptions hit a pandemic, a seasonal downturn, or a local crisis their income doesn't just dip; it vanishes. Why? Because their entire business exists in a logbook and their memory. There is no "business" to move online because they never had the tools, the literacy, or the confidence to go digital in the first place.

We looked at existing solutions and saw the pattern: every one of them was built for someone who speaks tech fluently. Every one of them assumed literacy, internet access, and comfort with screens. None of them were built for our women.

That's when we asked the question that changed everything: What if the barrier isn't the women it's the technology?

What It Does

Hagere Voice is an offline-first, voice-activated AI assistant designed specifically for non-technical women artisans to manage inventory, log transactions, and take control of their economic data entirely in their native language, Amharic.

Here's the magic: instead of navigating a dashboard, clicking buttons, or typing into forms, a woman simply speaks. She says:

  • "ሱሪ ሁለት ጨመር" → "Add 2 dresses to inventory"
  • "ቀሚስ አንድ ቀንስ" → "Remove 1 shirt"

The system understands, updates the database in real-time, and confirms the action. Every command is logged and auditable—creating an unbreakable digital record that proves her business's value and transactions. Over time, this data becomes her business resume—proof of her productivity and trustworthiness to lenders, partners, and customers.

How We Built It

We engineered a voice-first architecture that flips the traditional UI paradigm on its head.

The Stack:

  • Frontend: React + Vite (fast, responsive, lightweight)
  • Speech Recognition: Web Speech API with Amharic locale optimization (am-ET)
  • Natural Language Processing: Custom Amharic parser (amharicParser.js) that extracts intent, entity, and quantity from spoken input
  • Backend: Supabase (PostgreSQL + Realtime subscriptions)
  • Offline Resilience: Local-first caching with localStorage fallback
  • Deployment: Vercel (instant global CDN, fast cold starts)

The Intelligence Model:

We modeled voice command accuracy as a function of linguistic clarity and environmental noise. Our formula ensures reliable performance even in noisy market environments:

Accuracy = (Voice Clarity Score) / (1 + α × Background Noise Level) × 100%

Where:

  • Accuracy = intent recognition success rate
  • Voice Clarity = speech quality score (0–1)
  • Background Noise = ambient noise level in decibels
  • α = noise sensitivity coefficient (~0.02)

This informed our microphone threshold tuning and fallback UI (manual text entry) to ensure reliability in real-world conditions.

The Demo Flow:

  1. User taps ድምጽ ተናገር (speak) button
  2. Browser captures Amharic speech (am-ET locale)
  3. Custom parser extracts { action, itemName, quantity }
  4. apply_voice_command RPC updates inventory_items table and logs command to voice_commands (audit trail)
  5. Supabase Realtime pushes live updates to UI
  6. User sees before → after quantity change instantly

Offline Magic: If internet drops, commands queue locally. When connectivity returns, they sync automatically—zero data loss, zero friction.

Challenges We Ran Into

Linguistic Nuance Amharic has regional dialects, colloquialisms, and context-dependent meanings. A word spoken in Addis Ababa might be understood differently in Dire Dawa. We solved this by:

  • Training on region-specific voice samples
  • Building a flexible parser that handles synonym variations (ጨመር, ጨምር, አክል all mean "add")
  • Creating fallback flows so users can correct the system in real-time

User Trust Our biggest challenge wasn't technical—it was human. Women who've managed businesses on paper for decades were skeptical. "What if my data disappears? What if the machine gets it wrong?"

We addressed this through:

  • Transparency: Every command is logged and auditable; users can review their entire transaction history
  • Redundancy: Supabase backups + local cache = data is never lost
  • Feedback loops: Immediate visual confirmation ("Before: 5 dresses → After: 7 dresses") builds confidence
  • Real conversations: We sat with users, listened to their fears, and designed around them

Connectivity Reality Ethiopia's internet infrastructure is improving but remains inconsistent. We engineered for offline-first:

  • All voice processing happens in the browser (no server dependency for speech-to-text)
  • Command queue persists in localStorage
  • Automatic sync when connection restored
  • Progressive enhancement (fallback to manual input if microphone unavailable)

Accomplishments We're Proud Of

We successfully created an intuitive interface that removes the "tech barrier" for our users. By focusing on voice-first interaction, we ensured that a woman's economic potential is no longer limited by her ability to navigate complex digital platforms.

We built an audit trail that matters. Every voice command creates a timestamped, immutable record. For women who've never had "proof" of their business activity, this is revolutionary.

We shipped offline-first infrastructure. The app works without internet—and when it reconnects, it's bulletproof.

We localized AI responsibly. We didn't force English or generic AI models on our users. We built for Amharic speakers, not around them.

We created economic visibility. Over time, the data generated by this app becomes a woman's business resume proof she can show to lenders, investors, or new customers.

What We Learned

The most sophisticated AI is only as good as its accessibility.

We could have built a cutting-edge NLP model, trained on millions of data points, with 99% accuracy. But if a woman can't use it, it's worthless. We learned that the barrier to digital adoption isn't intelligence—it's friction.

We discovered that technology adoption isn't about features; it's about trust, speed, and language. When a woman speaks her native language and sees instant, tangible results, adoption happens naturally.

We also learned that the most impactful work happens at the intersection of empathy and engineering. Every design decision we made was informed by a real conversation with a real woman. Every line of code was a response to a real challenge she faces.

What's Next for Hagere Voice

Phase 2: Regional Expansion

  • Support for Tigrinya, Oromo, and Somali languages
  • Multi-language intent parser for cross-linguistic households
  • Regional dialect fine-tuning

Phase 3: Economic Multiplier

  • Direct integration with mobile payment systems (M-Pesa, Telebirr)
  • Customer relationship management (CRM) features for repeat clients
  • Simple lending portal that uses transaction history to help women access microloans

Phase 4: Community Ecosystem

  • Peer-to-peer marketplace for artisans to sell inventory and services
  • Analytics dashboard showing seasonal trends, bestselling items, and peak demand
  • Training modules (voice-driven) to help users upskill and expand their businesses

The Vision: Hagere Voice isn't just an inventory app. It's the first step toward economic agency for millions of women in East Africa who've been locked out of the digital economy by language, literacy, and the arrogance of designers who never asked what they needed.

In 18 months, we want Hagere Voice to be the trusted business partner for 10,000+ artisans, generating reliable business data that opens doors to credit, markets, and opportunity they've never had before.

Why This Matters

Numbers:

  • Ethiopia has 25+ million women-led micro and small enterprises
  • Only 5% have any digital record of their business activity
  • Women who can prove their business performance get 3x better access to credit

Our bet: If we give women tools built for them, in their language, they don't just adopt the technology—they own it, improve it, and teach others. That's not a feature. That's a movement.

Hagere Voice is our commitment to that movement.

Built With

Share this project:

Updates