Yojana Mitra

Inspiration

We kept hearing the same story from relatives in rural Telangana: eligible for pensions and farm support, but no one told them which schemes applied or which two documents to carry. English-only portals and rejected applications made it worse. We built Yojana Mitra so a citizen can speak for thirty seconds in Hindi or Tamil and walk away with a concrete, actionable list instead of another directory to read.

What it does

Yojana Mitra is a voice-first web app where you tap one mic button and describe your life in your language. It transcribes your speech, extracts eligibility signals like age, state, and occupation, matches you against 20+ curated central and state schemes, and returns ranked results with plain-language explanations, exact documents, and where to apply. It speaks the answer back and handles follow-ups like missing income certificates.

How we built it

We used Next.js 14 with Tailwind and shadcn/ui for a mobile-friendly frontend. API routes orchestrate Sarvam AI for STT, TTS, and chat, with Bhashini as a sovereign fallback across 22 languages. Scheme matching combines a hand-curated JSON database with rule-based scoring and optional LLM rewriting. API keys live in localStorage and travel via request headers - never stored server-side. Docker standalone output makes deployment one command.

Challenges we ran into

Indian-language ASR varies by accent, so we added a text fallback and cached demo transcripts for reliable judging. Eligibility rules are easy to hallucinate, so we constrained the LLM to our curated dataset and label every result as likely eligible, verify on official portal. We also had to get Devanagari and Tamil rendering right with Noto fonts from day one.

Accomplishments that we're proud of

A full under-60-second demo flow from voice to spoken pension recommendation with exact documents. Privacy-by-default with no sensitive data collection. Graceful demo mode when API keys are missing, so judges always see a working product. Shareable WhatsApp summary cards for CSC operators and family members.

What we learned

The awareness gap is the real bottleneck, not scheme availability. Voice beats forms for Tier 2/3 users, but text fallback is essential for demos. Sovereign Indian AI APIs like Sarvam and Bhashini are finally good enough for production-quality vernacular experiences at zero cost.

What's next for Yojana Mitra

Expand to all 22 scheduled languages via Bhashini, add more state scheme datasets, offline mode for ASHA workers, SMS summary delivery for users without smartphones, and integration with CSC appointment booking.

Built With

• bhashini
• docker
• next.js-14
• radix-ui
• sarvam-ai
• shadcn/ui
• tailwind-css
• typescript