Offline AI System Boosts Clinical Diagnosis in Low-Resource Settings

Access to specialized medical expertise is severely limited in many parts of sub-Saharan Africa, particularly in rural areas where physician-to-patient ratios are extremely low. Existing AI-powered diagnostic tools typically require stable internet connectivity and powerful hardware, making them impractical for frontline healthcare workers in these resource-constrained environments. To address this critical need, researchers have developed Aletheia. Aletheia is an innovative offline-first clinical decision support system specifically engineered for low-resource healthcare contexts. It is powered by Qwen2.5-3B-Instruct, a large language model that has been fine-tuned using Quantised Low-Rank Adaptation (QLoRA) on a comprehensive dataset of 27,000 clinical reasoning samples covering 50 prevalent East African disease conditions. This specialized training enables the system to provide accurate differential diagnoses. Evaluations demonstrate Aletheia's effectiveness, achieving an 80.0% Top-1 diagnostic accuracy and a 100.0% Top-3 accuracy across ten clinical case categories. Crucially, the system operates within a minimal memory footprint, peaking at approximately 3,630 MB RAM, well within the 7,168 MB budget of the Africa Deep Tech Challenge 2026. These results confirm the viability of deploying sophisticated LLM-based clinical reasoning directly at the primary care level, without reliance on cloud infrastructure, significantly enhancing diagnostic capabilities where they are most needed.