PRD: Humanness Image Collector

Objective: Build a scalable, mobile-first platform to collect 1000+ localized images and descriptions per village across India's 6 Lakh+ villages for AI model training. Target Quality: 98–99% data accuracy (first-pass approval rate >80%).

Assumptions

Contributors are field agents or volunteers : not necessarily tech-savvy
Minimum connectivity: 2G/3G with frequent offline periods in rural areas
Primary device: Android smartphones (iOS in V2)
NGO partners will handle ground-level onboarding of field agents
Admin team is internal JoshTalks staff (5-20 reviewers initially)

1. Problem Statement

Current AI vision models often fail to recognize India's unique cultural and geographic context - e.g., distinguishing a Durga Puja pandal from a regular tent, or identifying a handpump in rural Rohtak.

To make AI culturally inclusive of Bharat, JoshTalks AI needs:

1000 images per village across all Indian districts
Each image paired with an accurate text/voice description
Location-verified, fraud-resistant, high-quality data

The challenge: Build a tool that rural contributors can use with zero training, while ensuring data integrity at scale.

2. User Roles & Goals

Role	Primary Goal	Key Challenge
Contributor (Field User)	Capture & describe rural assets (handpumps, pandals, markets)	Low-network areas, varied literacy levels
Admin / Reviewer	Verify quality and monitor district-wise progress	Manual verification of massive datasets at scale