Case Study

Delivering Safe, AI-Powered HIV Testing With 98% Accuracy

Audere partnered with IndiVillage to train AI to interpret Rapid Diagnostic Tests used for detecting diseases like HIV, laying the groundwork for accurate, accessible healthcare diagnostics worldwide.

AUDERE

Audere, a global nonprofit innovating at the intersection of AI, mobile tech, and public health, partnered with IndiVillage to tackle a critical challenge of training AI to interpret Rapid Diagnostic Tests (RDTs) used for detecting diseases like HIV. While these cassettes offer life-saving insights, their visual variability of glare, lighting, and design differences makes automation difficult. IndiVillage brought precision and structure to Audere's raw data and laid the groundwork for accurate, accessible, AI-powered healthcare diagnostics worldwide.

631K+

interpretations completed

185K+

annotations processed

98%

accuracy achieved

Challenge

Making RDTs Machine-Readable

The project's complexity stemmed from the unpredictable nature of Rapid Diagnostic Test (RDT) images. Many were poorly lit, blurry, or overexposed, making it hard to distinguish result lines. Others included personally identifiable information (PII), requiring secure redaction. Inconsistent cassette formats and brand variations added further difficulty in creating uniform data for model training.

To complicate matters, a large volume of images lacked basic labeling. Cassette types had to be manually identified, and faint or unclear test lines demanded expert interpretation. These issues made automation alone insufficient, requiring a careful, human-led annotation approach to ensure precision, safety, and model reliability.

Solution

Expert-Led Annotation for AI Readiness

IndiVillage deployed a skilled team to annotate over 631,000 RDT images with both accuracy and speed.

The team visually assessed each cassette, classified test outcomes, and extracted essential metadata, including RDT brand, patient info, and batch IDs. Ambiguous or unclear cases were reviewed twice, and escalated to medical professionals when needed.

A rigorous feedback loop improved consistency over time, while low-quality or PII-containing images were flagged and filtered out. In total, more than 185,000 high-quality image annotations were processed with standardized tagging for downstream ML development. This expert-driven pipeline ensured clean, structured data, laying a strong foundation for AI models that could function reliably in real-world diagnostics.

Results

AI That Supports Real-World Diagnostics

Audere's AI models, fueled by IndiVillage's human-in-the-loop data pipeline, achieved up to 98% accuracy for certain RDT types in real-world scenarios.

The clean, well-tagged data enabled rapid AI deployment in diverse geographies, accelerating diagnosis, containment, and care, especially in underserved regions with limited clinical infrastructure.

This partnership helped Audere bring scalable, self-test interpretation to the global stage, bridging the gap between frontline healthcare needs and advanced AI capabilities. With IndiVillage's support, AI was no longer just a lab tool, it became a trusted decision-making ally for patients and health workers worldwide, delivering impact where it matters most.

“IndiVillage Tech Solutions has been a great partner in our mission to advance health equity through AI-enabled solutions. By providing accurately labeled images of the most widely used diagnostic tool globally, rapid diagnostic tests (RDTs), they have provided an invaluable input into our computer vision algorithms.”
Paul IsabelliChief Operating Officer, Audere