YojanaMitra: An AI-Assisted Framework for Government Scheme Eligibility Matching and Guidance

Abstract: Artificial Intelligence and Machine Learning are being utilised to automate eligibility assessment and provide personalisation in public service provision. They address longstanding issues faced by traditional systems such as fragmented scheme information, multi-condition eligibility criteria and limited citizen awareness. In this paper, we present YojanaMitra, a web application which provides access to 134 central and state government welfare schemes of India via a hierarchical and AI-driven matching pipeline. In the initial step, our matching pipeline uses a deterministic rule engine which supports both AND/OR logic of eligibility criteria to classify a citizen into three categories based on each welfare scheme: Eligible, Near Miss and Not Eligible. Only the schemes classified as Eligible are then ranked via a Random Forest classifier by engineering ten features from citizen-profile; finally, retrieval augmented conversational assistant provides citizen responses based on the matching scheme data alone. The application has been developed using Flask, SQLite database and Groq large language model API. It guarantees citizen security, structured profile-based matching and multilinguality. Automated testing shows accurate eligibility classification for all test cases; meanwhile, the accuracy of Random Forest ranking model is 0.597 with precision of 0.606, recall of 0.598, F1-score of 0.602 and ROC-AUC of 0.627 in the evaluation on held-out test set. We find that vulnerability and income margin are the two most important ranking features.

Keywords: Government Scheme Eligibility, Rule-Based Matching, Machine Learning Recommendation, Retrieval- Augmented Generation, Random Forest, E-Governance, Eligibility Prediction, Progressive Web Application, Flask, Multilingual Interface, User Engagement.