Voice-first, multilingual job matching for India's 450 million blue-collar workers. Speak in your language. Find work near you. No app download, no typing, no barriers.
India's informal sector employs over 450 million people. Most find work through middlemen who take hefty commissions, leaving workers underpaid and exploited.
India has 22 official languages and hundreds of dialects. Most job apps only support English and Hindi, excluding millions of workers from southern and eastern India.
A vast majority of blue-collar workers struggle with text-based interfaces. They need voice-first, conversational interactions in their own language.
Workers record a voice message in their native language. No typing needed. "Mujhe electrician ka kaam chahiye Bangalore mein" is all it takes.
Amazon Transcribe converts voice to text with support for 10 Indian languages including Hindi, Tamil, Telugu, Bengali, and more.
Amazon Bedrock (Nova Micro) extracts structured data from natural speech: job type, location, salary expectations, skills, and availability.
Multi-dimensional matching algorithm scores jobs on location proximity, skill relevance, salary fit, and schedule compatibility. Returns the top matches.
Results are formatted and delivered in the worker's native language with contact details, salary info, and location. The full conversation is maintained across sessions.
6 serverless functions: Message Router, Voice Processor, Entity Extractor, Matcher, Response Generator, Web Chat
3 tables with GSIs: Sessions (TTL), Job Postings (location + category indexes), User Profiles
Nova Micro model for entity extraction and multilingual response generation via inference profiles
Real-time speech-to-text in 10 Indian languages with confidence scoring
Two REST APIs: WhatsApp webhook (EUMS) and Web Chat endpoint with CORS & throttling
Cross-region fan-out from us-east-1 (EUMS) to ap-south-1 (Lambda). Dead Letter Queue for reliability
Voice file staging with auto-cleanup lifecycle policy. Static website hosting for the web app
HTTPS on custom domain kaamconnect.rahulsingh.xyz with SSL certificate via AWS Certificate Manager
End User Messaging Social for WhatsApp Business API integration in us-east-1
4 stacks in Python: Foundation, API, Processing, Monitoring. Full infrastructure as code
Custom dashboards, Lambda log retention, error tracking, and performance monitoring
Least-privilege roles per Lambda. Fine-grained Bedrock, Transcribe, and DynamoDB permissions
Integrated AWS End User Messaging Social in us-east-1 with Meta WhatsApp Business API. Full inbound message pipeline working.
Inbound WorkingMessages flow from EUMS (us-east-1) via SNS to Lambda functions in ap-south-1 (Mumbai) for processing. Fully serverless.
Architecture ReadyOutbound messaging requires Meta Business Verification which takes 2-7 business days. With the hackathon deadline approaching, we couldn't wait for approval.
Blocked by VerificationBuilt a full web chat interface reusing the same backend pipeline. Same AI, same matching, same multilingual support. Zero code waste.
Live & WorkingThe WhatsApp integration is not vaporware. Every component is deployed and tested:
One tap to record. Speak naturally in your language. No keyboards, no forms, no confusion. Just a conversation.
Amazon Bedrock extracts job type, location, salary, and skills from natural speech. "Mujhe driver ka kaam chahiye" is enough.
Not a form. Not a search bar. A conversation. KaamConnect asks follow-up questions, clarifies details, and guides workers through matching.