Gestational Diabetes Mellitus (GDM) and Excessive Gestational Weight Gain (EGWG) pose significant health risks to mothers and children in Bangladesh. Although mobile health (mHealth) apps offer a promising avenue for managing these conditions, generic solutions often fail due to a lack of cultural and contextual adaptation. This study presents a user-centred approach to define the essential features of an Artificial Intelligence (AI)-integrated mHealth app specifically for Bangladeshi women. We conducted a mixed-methods study, beginning with a survey of 15 pregnant or postpartum women to understand their digital literacy, health information behaviours, and needs. Following this, we conducted a focus group discussion to deepen their understanding of their experiences and expectations. Using the six-phase thematic analysis framework by Braun and Clarke, we analyzed qualitative data to identify core user requirements. Our analysis revealed five key themes: (1) accessibility and usability, (2) cultural and contextual relevance, (3) core health management features, (4) trust, privacy, and security, and (5) social and community support. These findings form the basis of a conceptual model and corresponding low-fidelity prototypes for a culturally-attuned, accessible, and trustworthy mHealth app. This research provides actionable insights for developing solutions that are not only clinically effective but also resonate with the lived realities of users in low-and middle-income countries (LMICs), such as Bangladesh.