How It Works

TruVector's mobile workflow collects African-context data with clear consent, automated quality checks, structured validation, and traceable delivery. Every data point includes comprehensive metadata for responsible AI development.

1. Onboarding and Identity

Users create accounts using phone-number authentication, providing basic context for dataset representativeness: region, primary languages, device type, and optional demographic data (with explicit consent). Granular consent is captured per data type, clearly explaining what will be collected, why it's needed, and how it will be used.

2. Data Capture

Users access a task feed with clear instructions, examples, and quality requirements. The platform supports multiple data types:

Audio

Speech, read-aloud, conversational snippets, pronunciation variants

Video

Controlled prompts, environment samples, scripted phrases

Images

Objects, scenes, context-specific visuals

Text

Messages, code-switching examples, local phrasing

Surveys

Structured metadata and questionnaires

In-app guidance helps users submit high-quality data: quiet environments for audio, stable framing for video, and privacy reminders.

3. Automated Quality Checks

Before acceptance into validation, automated checks flag issues: audio clipping/distortion, video resolution/stability, image blur, and text spam patterns. Users receive immediate feedback and can resubmit if needed.

4. Structured Validation

Multi-tier validation ensures quality: peer reviewers confirm basic correctness, higher-trust reviewers handle complex tasks, and expert reviewers validate linguistic/cultural context. Golden tasks calibrate reviewer accuracy. Confidence-weighted decisions route uncertain samples for additional review.

5. Rewards and Compensation

Users earn credits for validated submissions and eligible review work. Rewards are delivered through scalable mechanisms: mobile data bundles, airtime, vouchers, or other region-appropriate options, configured based on dataset requirements.

6. Dataset Delivery

Licensed datasets are delivered with validated samples and comprehensive metadata per data point:

Sample data

Validated audio/video/image/text

Rich metadata

Language, region, consent records, age, phenotypical data

Quality indicators

Automated check results and validation outcomes

Provenance

Complete audit trail and validation history

Documentation & licensing

Dataset scope, limitations, intended use, and licensing terms

Outcomes

This workflow produces datasets that enable AI systems to:

Perform better on African languages and accents

Generalize to real African environments

Reduce bias from thin or skewed data

Deploy responsibly with traceable consent and validation