Haqdar (हक़दार)

Rightful Claimant

Haqdar is an AI-powered welfare discovery tool built for rural and low-income Indians. Users describe their situation freely in Hindi — by typing or speaking — and Haqdar tells them exactly which government schemes they qualify for, what documents they need, and where to apply. No English required. No middleman. No prior knowledge of what schemes even exist.

Under the hood, Haqdar runs a semantic search engine over 1,941 real government schemes scraped from MyScheme.gov.in, retrieving the most relevant schemes for a user's profile and reasoning over their actual eligibility text using the Anthropic Claude API. The result is a set of plain-Hindi scheme cards with document checklists and action plans — delivered in under two minutes.

Built at the Anthropic Hackathon, IIT Delhi, April 2026.

The Build System — PACT

Parallel AI Collaboration with a Tether

Haqdar was built by a team of four complete beginners in under three hours using a coordination protocol designed from scratch called PACT.

The core insight behind PACT is simple: AI agents are excellent at building individual components but blind to each other. Left uncoordinated, four agents building in parallel will produce four pieces that don't fit together. PACT solves this by introducing a dedicated Integrator — a human operator who coordinates a central orchestration AI, manages the shared codebase, and owns integration across all checkpoints — while three Builders each own exactly one component and never touch anything else.

Checkpoints are task-driven, not time-driven. At each checkpoint the Integrator pulls all Builder work, tests specifically whether components are communicating correctly, and fixes only the connection points before releasing the next round of build prompts. Builders receive updated memory files so their AI always builds on accurate shared assumptions.

No team member wrote a single line of code manually. The Integrator used a coding agent with full codebase visibility. Builders used chatbot tabs and created files manually in VS Code. A persistent Integration Claude instance maintained the architectural overview across the entire build.

PACT was conceived and designed by the team lead, a student with no prior software development experience.