Scheme Sathi

Inspiration

Growing up in India, I watched family members miss out on government benefits they were fully entitled to — not because they didn't qualify, but because nobody told them they qualified. The Indian government runs over 300 welfare schemes, but awareness and accessibility remain massive barriers, especially for rural citizens, farmers, and first-generation smartphone users.

I built Scheme Sathi ("Scheme Companion") to solve exactly this problem: give every Indian citizen a personal AI advisor that speaks their language, understands their life situation, and instantly surfaces every government benefit they deserve.

What It Does

A citizen fills a simple 12-field profile form. Scheme Sathi then makes 3 chained calls to ASI-1 Mini and returns:

• ✅ Every government scheme they qualify for (from a database of 40 real schemes)
• 📊 Eligibility status and confidence score for each scheme
• ⚡ A personalized action plan with exact steps to apply today
• 📋 A printable document checklist generated by a 4th ASI-1 call
• 💬 A conversational chat to ask follow-up questions — with full context

How I Built It

The entire app is a single HTML file — no frameworks, no build tools, no backend. It runs entirely in the browser.

The core innovation is the 3-Step Chained Prompting Strategy using ASI-1 Mini:

Call 1 — Profile Categorization ASI-1 classifies the user into structured metadata: { "income_bracket": "BPL", "primary_demographic_groups": ["rural-farmer", "SC"], "vulnerability_score": 9, "key_needs": ["agriculture", "health", "housing"] }

Call 2 — Eligibility Matching The categorization from Call 1 is fed back as context alongside all 40 schemes. ASI-1 reasons over each scheme's eligibility criteria and returns confidence scores, eligibility status, and benefit descriptions.

Call 3 — Action Plan The top 5 matched schemes feed into a final call that generates a warm, personalized action plan — with steps, documents to gather, and encouragement — in the user's chosen language (English or Hindi).

Call 4 — Document Checklist An on-demand 4th call generates a per-scheme document list, rendered as a clean printable HTML page that opens directly in a new tab.

Challenges I Faced

1. Chained API + Browser Security Browsers block window.open() calls made after an await. The checklist feature needs to open a new tab and write AI-generated content into it. The fix was to open the tab synchronously on click (before any await), show a loading spinner immediately, then write the final HTML into the already-open tab once the API responded.

2. JSON Parsing from LLM Output ASI-1 sometimes wraps JSON in markdown code blocks. I wrote a extractJSON() fallback parser that strips ```json fences, then tries bare object/array extraction before throwing — ensuring graceful degradation even on imperfect responses.

3. Bilingual UI Without a Framework Supporting English and Hindi across the entire UI — including dynamic AI responses — without React or any framework required a clean TRANSLATIONS object and careful application of language context to every ASI-1 prompt.

4. Token Efficiency Sending all 40 scheme objects in Call 2 risks hitting token limits. I solved this by sending only the compact {id, name, category, eligibility} subset to ASI-1, then merging the AI's output with the full static scheme data on the client side for rendering.

What I Learned

• Chained prompting is fundamentally more powerful than a single prompt. Each ASI-1 call in Scheme Sathi builds on the previous one — the categorization primes the eligibility matching, which primes the action plan. The output quality is dramatically better than any single-shot approach.
• ASI-1 Mini has strong reasoning on structured eligibility rules — it correctly handles edge cases like age-gated schemes, BPL-only schemes, and gender-specific programs without explicit rule-coding.
• Static sites can do a lot. A single HTML file with vanilla JS, zero dependencies, and a good API is genuinely enough to build something impactful.

What's Next

• Integrate with DigiLocker for one-tap document verification
• Add voice input for low-literacy users
• Expand to 200+ schemes with state-specific programs
• Partner with CSC (Common Service Centres) for assisted-mode kiosks in rural areas

Built With

• asi-1
• blob
• chrome
• css3
• html5
• javascript
• localstorage
• mini
• share
• vanilla
• web