Claiming the unclaimed: Turning government schemes into guaranteed rights
Today, millions of Indians miss out on life-changing welfare simply because they don't know they are eligible
The Inspiration: Closing the "Awareness Abyss"In Bharat, a "Right" is often a "Secret." Every year, billions in welfare remains unclaimed—not because of a lack of funds, but because of a lack of awareness. We realized that for a farmer in Satara or a tribal family in Odisha, the gap between a government scheme and their actual bank account is a 100-page PDF written in a language they don't speak.We were inspired to flip the script: Don't make the citizen search for their rights; make the rights search for the citizen. We envisioned Adhikar-Aina, a "Sovereign Rights Mirror" that reflects a citizen's eligibility before they even think to ask.
How We Built It: The Intelligence LakehouseWe leveraged the Databricks Data Intelligence Platform to move from "Static Data" to a "Proactive Pulse."The Medallion Foundation: We ingested messy government gazettes and scheme notifications into a Bronze Layer using Databricks Autoloader, ensuring we could handle high-velocity updates from across 22 states.Semantic Logic Extraction: Using Databricks AI Functions, we transformed dense legal jargon into a structured Silver Layer. We moved beyond keyword matching to true logical extraction.The Citizen Twin: We modeled citizen data as "Digital Twins" in the Gold Layer. This allowed us to perform continuous, real-time matching between personal attributes and scheme requirements.The Interface: We built a multilingual voice interface using Streamlit (hosted as a Databricks App), connected to Databricks Vector Search for semantic retrieval that understands context, not just text.We utilized Claude Code with the MCP (Model Context Protocol) to "vibe code" this entire infrastructure, allowing us to manage Unity Catalog permissions and deploy complex pipelines directly from our terminal in minutes.
The Challenges: Turning Law into MathThe hardest part was translating "Legal Language" into "Boolean Logic." Eligibility isn't just a keyword; it’s a rigorous set of constraints. We modeled the matching probability P as a function of the Citizen profile C and the Scheme rules R:P(Eligibility) = {i=1}^{n} (C_i, R_i) Where C_i represents citizen attributes (income, land, age) and R_i represents the statutory constraints.Ensuring that our LLM didn't "hallucinate" a right was critical. To solve this, we used Unity Catalog Lineage to trace every AI-generated recommendation back to the original, official government PDF, providing a "Source of Truth" that builds user trust.
What We Learned : We learned that Sovereign AI isn't about the biggest model; it’s about the best Data Governance. By using Databricks Unity Catalog, we proved that we could run complex welfare simulations while keeping PII (Personally Identifiable Information) locked behind "Clean Rooms."Most importantly, we saw that with the right AI tools, a single developer can build what used to take an entire government department months to coordinate. Efficiency is the new scale.
Log in or sign up for Devpost to join the conversation.