SahayakAI

Inspiration

In India, millions of citizens, especially farmers and rural households struggle to access government welfare schemes due to language barriers, complex forms, and lack of awareness. We were inspired by the idea of building a digital extension agent that bridges this gap using Agentic AI.

What it does

SahayakAI is an agentic AI system designed to improve access to citizen services and welfare schemes in India. It enables users to ask questions in natural language (text or voice) about government subsidies, schemes, or grievance processes. SahayakAI integrates Granite large language models for natural language understanding, a Retrieval-Augmented Generation (RAG) pipeline to ensure responses are grounded in official documents, and IBM’s Agent Development Kit (ADK) for workflow automation. Together, these components enable the system to provide accurate responses, facilitate form completion, assist in filing complaints, and monitor application progress. The system further aggregates citizen interactions into dashboards for officials, helping them detect common issues such as subsidy delays or shortages. With multilingual support, OCR for documents, and secure API integration, SahayakAI bridges the gap between government services and citizens, increasing transparency, accessibility, and efficiency.

How we built it

• Data Preprocessing & Knowledge Base: Collected public government policy documents and scheme eligibility criteria. Built a searchable RAG pipeline.
• Citizen Query Handling: Used Granite LLM to parse natural language queries (text/voice).
• Workflow Automation: Leveraged IBM ADK to orchestrate tasks - fetching eligibility, filling out application forms, submitting grievances.
• Multimodal Support: OCR integration so citizens can upload paper documents for verification.
• Insights Dashboard: Aggregates helpline and chatbot interactions, leveraging Agentic AI to autonomously detect emerging issues (e.g., fertilizer shortages, subsidy delays), prioritize them, and suggest proactive interventions.

Built With

• granitellm
• ibm-adk
• ocr
• python
• rag
• torch
• vector-db