In rural India, doctor-to-patient ratio is estimated at 1:1500, well below the national average. Connectivity and access to modern equipment is a further challenge. Rapid developments in artificial intelligence (AI) could help frontline medical personnel, if adapted to these resource constrained environments. This paper evaluates algorithms that could form the basis of lightweight AI models supporting primary diagnosis in settings characterized by low computing power and poor connectivity. Using the publicly available UCI Heart Disease dataset comprising 303 patient records with 14 clinical features, four algorithms: Logistic Regression, Decision Tree, Random Forest and k-Nearest Neighbors (KNN), were evaluated for accuracy, precision–recall, average precision (AP) and file size. Deployment feasibility was assessed based on storage footprint and inference efficiency. As per the results, KNN, with AP of 0.927, Accuracy of 90.16% and File Size of 57.87 KB strikes the right balance between accuracy and file size and can be developed further. The other three algorithms delivered results with lower accuracy, suggesting they can be more appropriate for screening, risk stratification or early warning systems. The use of the Kaggle-hosted Jupyter Notebook served as a reasonable simulation of algorithmic performance in a low-compute context as no GPU acceleration was used. Using simple data with no images and measuring model file size further confirmed fit to the operational challenges of basic hardware. Next steps include piloting on target hardware using larger region and disease specific datasets after securing required legal and ethical clearances.