In India’s informal and blue-collar employment sectors, millions of workers face fraudulent job listings, unfair wages, and lack of social security access. Simultaneously, local employers struggle to find vetted, reliable trade talent. We built KaamSetu to bridge this trust deficit by providing a voice-enabled, scam-resistant hiring and social security verification platform tailored for every worker, regardless of digital literacy or connectivity.

What it does:
AI Proximity & Fit Matching: Connects job seekers with verified local vacancies using a transparent Gemini AI Fit Score (factoring 40% distance, 30% skills match, and 30% employer trust).

EPFO & ESIC Employment Verification: Evaluates employer continuity and worker trade readiness using consent-driven government social security signals without exposing private identifiers (UAN/Aadhaar masking).

AI Scam Scanner & Reporting: Leverages Gemini multimodal capabilities to analyze job postings and scam evidence (via file upload or live camera photo capture) with direct escalations to Cyber Crime (1930) and Labour (14434) helplines.

Offline-First Resilience: Built as a Progressive Web App (PWA) with local action caching, enabling workers to apply or save walk-in interview slots even with zero connectivity, syncing automatically once reconnected.

Multilingual & Voice-First Access: Full speech navigation and Hindi/English voice dictation via Web Speech API, plus instant access to government welfare schemes (PM-SYM, eShram, Ayushman Bharat).

How we built it:
AI & Intelligence: Powered by the Google Gemini API (Gemini 1.5 Flash) for automated employer verification, fraud confidence scoring, resume parsing, and fit scoring.

Frontend: HTML5, CSS3, JavaScript (ES6+), Leaflet.js with OpenStreetMap for real-time proximity mapping, Web Speech API for voice interactions, and Service Workers (sw.js) for offline caching.

Backend & Storage: Python 3.10+, FastAPI, Uvicorn, and Pydantic with a persistent JSON-backed database layer.

Challenges we ran into :

1. Handling seamless offline synchronization so applications, complaints, and profile edits queue reliably without data duplication when internet access is spotty.

2)Tuning Gemini prompts to output reliable structured fraud-risk confidence scores and balanced match percentages across varied trade skill sets.

3. Ensuring zero sensitive PII leakage when validating government welfare and social security markers (EPFO/ESIC).

What’s next:
Integrating direct SMS/USSD application gateways for basic feature phones.

Expanding vernacular voice AI models for additional regional dialects and automated interview pre-screening.