Skip to main content
AboutProblemsGuidelinesTerms & conditions

Problem Statement 5

"Niti-Setu" - Voice-Based Scheme Eligibility Engine​

Focus: Voice Interface | RAG (Retrieval-Augmented Generation) | Document Processing | Rule-Based AI

1. Background:​

Government launches 100+ agricultural schemes annually with combined budgets of ₹50,000+ crores. However, adoption rates stay below 30% because eligibility criteria are buried in complex 50-page PDFs. Farmers can't decode bureaucratic language to check if they qualify.

2. Current Problem:​

1. PDF Bureaucracy

  • Scheme guidelines written in technical English/Hindi
  • Eligibility rules hidden in Section 4.1.2, subsection (c)
  • Impact: Farmers miss benefits they're entitled to

2. Generic Chatbots

  • Provide generic answers, not personalized eligibility
  • No proof or citations from official documents
  • Impact: Farmers don't trust the advice

3. Manual Form Filling

  • Application forms require reading entire guidelines
  • No guidance on required documents
  • Impact: 40-50% applications rejected due to errors

3. Goal:​

Build a consultant AI that converts PDF bureaucracy into personalized Yes/No eligibility decisions with proof.

Beyond chatbots → Create a platform that:

  1. Ingests official scheme PDFs automatically
  2. Analyzes farmer profile against eligibility rules
  3. Provides binary decisions with document citations
  4. Simplifies application process

4. Expected Solution (MVP Requirements):​

1. Voice-Enabled Profile Input​

Capture farmer details via:

  • Voice Input: Speech-to-text in Hindi/English
  • Form Fallback: Manual entry option
  • Data Collected: State, district, land holding (acres), crop type, social category

Output: Structured farmer profile

2. RAG Engine (AI Backend)​

Process scheme documents:

  • PDF Ingestion: Pre-process 2-3 real scheme PDFs (PM-KISAN, PM-KUSUM, Agri-Infrastructure Fund)
  • Vector Search: Find "Eligibility Criteria" sections
  • Profile Matching: LLM compares farmer data vs PDF rules
  • Citation Extraction: Identify exact paragraph/page supporting decision

Demo Goal: Input profile → Get "Eligible/Not Eligible" with PDF proof in 10 seconds

3. Proof Card Generator​

Display results clearly:

  • Eligibility Status: Visual card (✓ Eligible for ₹6,000/year)
  • Document Proof: Screenshot or text snippet from PDF
  • Citation: "Page 4, Paragraph 3: 'All farmers with landholding less than 2 hectares...'"
  • Next Steps: List required documents for application

Validation: Test with 10 real farmer profiles, verify accuracy against manual PDF reading

4. User Dashboard​

Real-time interface:

  • Profile Summary: Saved farmer details
  • Scheme Cards: List of applicable schemes
  • Proof Display: Expandable citations from PDFs
  • Impact Metrics:
    • Schemes analyzed
    • Eligibility checks performed
    • Average response time

5. "Level Up" Features:​

Advanced Features (Choose 2-3)

  • Auto Form Filler: Pre-fill application PDFs using pdflib
  • Multilingual TTS: Read results in Hindi/Marathi/Tamil
  • Document Checklist: Auto-generate required documents list
  • Scheme Comparison: "You qualify for 3 schemes, here's the best one"
  • Application Tracker: Monitor submission status

6. Tech Stack:​

Core

  • Backend: Node.js/Express
  • Database: MongoDB (vector storage for embeddings)

AI/ML

  • Orchestration: LangChain
  • LLM: OpenAI API / Gemini / Llama
  • Embeddings: text-embedding-ada-002 or sentence-transformers
  • Speech-to-Text: Web Speech API / Google Cloud Speech

Frontend

  • Framework: React.js
  • Voice Input: Browser SpeechRecognition API

External APIs

  • Government scheme PDFs from official portals