What it is

Ardham (అర్థం) is an on-device, offline-first voice verification system that ensures rural and visually impaired borrowers truly understand their loan terms before signing.

The Problem

Rural borrowers and self-help group women often sign 10+ page loan agreements they cannot read. Field officers collect signatures or thumbprints with zero proof of comprehension, leading to hidden fee deductions and predatory 36%+ debt traps.

How it works

1. Camera Scan: Uses Google Gemma 4 vision to read 1 to 10+ pages of a loan card or Key Fact Statement (KFS).
2. Zero-Hallucination Guard: Recomputes all math in Python and locks every number to the printed card so the model never invents figures.
3. Plain Mother Tongue Speech: Explains terms in spoken Telugu and Hindi without technical jargon.
4. Lakshmi AI Voice Avatar: A real-time lip-sync avatar allowing blind borrowers to navigate by voice (“Next”, “Back”, “Repeat”) and answer questions hands-free.
5. Digital Verification Receipt: Generates an SHA-256 sealed digital receipt and dynamic QR code for audits and WhatsApp sharing.

Built with

Google Gemma 4 (on-device Ollama SLM + Gemini API fallback), Python, FastAPI, React, TypeScript, and Agent Skills. 100% offline-ready, open-source under Apache 2.0, and deployed on Render.