Jobhunt — Voice-First Blue-Collar Hiring Platform

India's blue-collar labour market is broken for both sides. Skilled workers — electricians, welders, fitters, masons — have no way to prove their skills beyond word-of-mouth. Employers can't find and verify candidates quickly. Middlemen take cuts. Everyone loses.

I have built Jobhunt to fix this. Workers complete a 3-step voice onboarding in Telugu, Hindi, or English — they speak their trade and experience, and the AI extracts and pre-fills the form. The result is a verified Skill Passport: a shareable public profile with a trust tier, endorsed skills, and a match score against every open job.

Employers post jobs in under 10 seconds (with an AI-assisted description), search candidates ranked by a transparent 5-factor match score (Skills × Distance × Experience × Wage × Trust), and manage their hiring pipeline on a Kanban board. Every stage transition sends a real-time notification to the worker.

We built it on Next.js 16 with the App Router, Prisma ORM on Supabase (PostgreSQL), NextAuth for JWT-based auth (with email/password and demo accounts), and Zod for shared client-server validation. The SmartMatch algorithm pre-computes scores on job post and worker onboard, making the feed sub-second.

The hardest challenge was building a system that degrades gracefully — the AI voice flow, LLM job descriptions, and real-time notifications all have deterministic fallbacks so the app works 100% without any external service configured.

What I learned

That "AI-first" doesn't mean "AI-dependent". Every AI feature we built — voice onboarding, job description generation, match scoring — has a fully functional deterministic fallback. We learned to design the AI layer as an enhancement, not a dependency. We also learned how much the right data model matters: pre-computing match scores into a cache table made the difference between a 3-second feed and a sub-second one.