She has cooked for 200 weddings. That is a skill. Give it a certificate.

Live site: https://heerforit.github.io/HUNAR

India has 100 million informal workers — cooks, tailors, weavers, caregivers, security guards — with decades of real skill and no credential to prove it.

No certification system was ever designed for them.

The ones that exist require written tests, physical travel, and weeks of waiting. Workers fail not because they lack skill — but because the system was never built for them.

I found this problem fifty metres from my hostel room in Kota. My warden has cooked professionally for fifteen years. She has never received a single certificate for any of it.

The credential gap is not a competence failure. It is a design failure.

HUNAR certifies workers through three layers — on any basic Android phone, in four minutes, at zero cost to the worker:

Google Gemini evaluates all three together and generates a score breakdown across criteria including Technique, Timing, Hygiene, Ingredient Handling, and Presentation.

Result: a QR-verified digital certificate the worker owns permanently — shareable on WhatsApp to any employer instantly.

Four people. Four different skills. One identical answer when asked if they had a certificate for their work:

No.

That answer is why HUNAR exists.

This project went through two complete versions based on real user feedback. Both versions are preserved in this repository to show the iteration process.

File: Hunar3.aia

Built the initial prototype using MIT App Inventor — a block-based visual programming environment. The goal was to prove the concept worked before investing in a full production build.

What it had:

• 5 screens — language selection, skill selection, voice assessment, processing, certificate
• TextToSpeech component reads questions aloud in worker's language
• SpeechRecognizer captures spoken answers — no typing required
• TinyDB with shared namespace HUNARdb persists data across all 5 screens
• Sharing component sends certificate via WhatsApp
• Questions loaded dynamically based on skill — one assessment screen handles all 5 skills

What changed after testing with my warden:

• She kept looking at the question text even though the app was reading it aloud → made question text smaller, microphone button much larger
• She did not understand "Level 2" → added the word "Proficient" next to the number

Shortcomings identified:

• Questions were in English only — contradicted the multilingual design principle
• Any spoken answer scored a point — no real AI evaluation of answer quality
• Certificate was text only — no actual QR code generation
• Limited to 5 skill categories

Decision: These shortcomings were too fundamental to patch. Rebuilt from scratch in Flutter.

File: Hunar (3).zip

Complete rebuild in Flutter for production quality, addressing every shortcoming identified in V1.

What was added:

• Video demonstration assessment — worker records herself working on camera
• Google Gemini AI evaluates video across 7 criteria with percentage scores
• Employer voice feedback — 30-second recorded comment, no typing
• 14 profession categories (up from 5)
• Two-tier certification — Basic (3 tests) and Advanced (6 tests)
• Employer marketplace — browse, filter, and contact verified workers
• Full multi-lingual support via LocalizationService
• flutter_tts for voice-first accessibility
• shared_preferences replaces TinyDB for cross-screen persistence
• share_plus for WhatsApp certificate sharing
• Demo mode with 11 pre-loaded workers for judges

AI Analysis output (real result):

git clone https://github.com/heerforit/HUNAR.git
cd HUNAR
unzip "Hunar (3).zip"
cd "Hunar (2)/Hunar"
flutter pub get
flutter run -d <device-id>
flutter build apk --release

Requirements: Flutter SDK 3.7.2+, Dart 3+, Android SDK 21+

Demo login: +919876543210 through +919876543220, password: demo123

1. Go to ai2.appinventor.mit.edu
2. Projects → Import project (.aia) → select Hunar3.aia
3. Connect → AI Companion → scan QR code with MIT AI2 Companion app on Android

lib/
├── main.dart
├── models/
│   ├── user.dart
│   ├── professions.dart
│   └── challenges.dart
├── screens/
│   ├── landing_page.dart
│   ├── signup_screen.dart
│   ├── login_screen.dart
│   ├── profession_selection_screen.dart
│   ├── challenge_screen.dart
│   ├── video_recording_screen.dart
│   ├── analysis_screen.dart         ← AI scoring results
│   ├── profile_screen.dart          ← Verified certificate
│   ├── marketplace_screen.dart      ← Employer browsing
│   └── worker_detail_screen.dart
├── services/
│   ├── localization_service.dart    ← 5 Indian languages
│   ├── tts_service.dart             ← Text-to-speech
│   └── app_data.dart
├── widgets/
└── utils/

HUNAR earns from employers and government — never from workers.

Worker pays: Rs.0. Always.

Break-even: 500 assessments per month. Gross margin: ~95%.

1. Voice recordings deleted within 72 hours of assessment
2. AI defaults to higher score in ambiguous cases — worker never penalised
3. Voice consent required before every session in worker's language
4. Worker told what the certificate means and how to delete their record
5. AI evaluates technique only — not camera quality, environment, or presentation quality

• Technovation Girls 2025–2026 — Senior Division
• Global Appathon 2026 — MIT App Inventor Foundation — Youth Individual (13–17)

Raghunandan Gupta — Senior Computer Scientist, Adobe Systems, Noida. Business mentorship, revenue model validation, go-to-market feedback. Former Technovation Girls judge.

Hostel Warden, Kota — Primary user research participant. Her fifteen years of professional cooking and beauty work with zero credentials directly shaped every design decision in both versions of HUNAR.

Mess Cook, Security Guard, Gym Trainer, Kota — User research participants. Confirmed the credential gap across four different skill categories.

Heer — Class 11 student, Kota, Rajasthan, India

Built HUNAR while preparing for JEE, living in a hostel fifty metres away from the woman the app was designed for.

Invisible expertise becomes economic power.

This is HUNAR.