| The lack of timely and reliable maternal and foetal health monitoring remains one of the most significant challenges in resource-constrained settings. The traditional health monitoring systems have a very strong architectural dependency on the cloud, which results in high latency, the need for a continuous internet connection, and serious data privacy concerns. To address the challenges mentioned above, this research proposes an integrated Edge AI and Conversational AI approach that can provide rapid and privacy-preserving maternal and foetal health monitoring. This system integrates a multi-output Random Forest for parallel prediction, a fine-tuned TinyLlama chatbot, and a Bayesian Network for personalized recommendations. The Random Forest model achieved an accuracy of 96.47% for CTG classification and 90.8% for maternal risk prediction, with consistently high precision, recall, and F1-scores across risk categories. To achieve efficient edge deployment, TinyLlama was domain-adapted to maternal healthcare questions and quantized to 3-bit precision, effectively reducing the model size while maintaining response quality and conversational coherence(perplexity 4.48; cosine similarity 0.86). The correlations between the most significant maternal factors are represented by the Bayesian Network, which enables probabilistic conclusions and customized recommendations that consider chatbot responses according to each user profile. The suggested system provides a promising scalable and low-latency solution that can be used to improve maternal and foetal healthcare in resource-constrained environments by utilizing edge computation, contextual intelligence, and interpretability. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.