Entitled
A voice agent that finds every government scheme a family is owed, fills the form and tracks it till the money lands, only on their yes.
Built at AgentX Hackathon 2026
Team Harshi's Team: Harshi Singh, Gaurav Tiwari
Built with Python, FastAPI, LangGraph, JSON-Logic, Playwright, SQLite/PostgreSQL, Next.js, TypeScript, PWA, ONNX Runtime Web, UniOCR, Whisper, Render, Vercel
About the project
Problem: Millions of Indian families miss welfare schemes they are legally entitled to. The rules are scattered, the paperwork is confusing, and nobody follows up after an application is sent. Schemes are a citizen's right, not charity.
Solution: Entitled is a voice-first agent for the family and for the Gram Saathi operator (ASHA, CSC or panchayat worker) who helps them. The family speaks in their own words, and one sentence can answer many questions. Entitled checks all 16 schemes and explains every result with a rule check and its official source. It points out near misses ("one income certificate away") and clashes, where only the higher of two pensions counts. Missing documents become one ranked to-do list. Then it fills the application and follows it until the money arrives.
How it works: A LangGraph agent works in six stages (interview, eligibility, documents, form, confirm, track) and uses nine tools. It remembers each household across visits. The LLM never decides eligibility: deterministic JSON-Logic rules do, so an unknown fact stays undecided instead of being guessed. Documents are read on the phone with UniOCR, and only the last 4 digits of an Aadhaar number leave the device. Speech is transcribed on the phone with Whisper. Nothing is submitted without the person's own plain "yes". This is enforced in code with a signed, one-time confirmation token: "No, don't send it yet" never sends anything. Operator mode shows every household's next task and overdue follow-ups, with an action log that masks personal numbers.
Impact: For our demo household, Kamla Devi, Entitled found and applied for ₹18,000 a year, a ₹1.2 lakh housing grant and ₹5 lakh of health cover, and tracked them to the first instalment. It runs with no API keys, has 171 automated tests, and scores 400 out of 400 on an eligibility evaluation.