William Niemiec
Back to research
Using LLMs to Generate mHealth Applications

Regional School of Machine Learning and Artificial Intelligence (ERAMIA)

Using LLMs to Generate mHealth Applications

William Niemiec, Anderson Rocha Tavares, Érika Cota

Published in
Regional School of Machine Learning and Artificial Intelligence (ERAMIA)
Year
2026
DOI
10.5753/eramiars.2026.35112

Abstract

The development of mobile health (mHealth) applications is resource-intensive, and the effectiveness of Large Language Models (LLMs) in automating this process remains understudied. This paper evaluates a Multi-Agent LLM Framework (MALFRA) pipeline that translates natural-language specifications into executable code for four mHealth applications. Results show that the framework produced compilable applications with relatively few statically detected bugs and no vulnerabilities detected by the configured static-analysis rules. However, functional completeness was limited: 46.38% of the evaluated features were not implemented. MALFRA particularly struggled with complex multi-step workflows and backend-dependent functionality, indicating that successful code generation and compilation do not imply functional completeness for mHealth applications.

How to cite

William Niemiec, Anderson Rocha Tavares, Érika Cota. Using LLMs to Generate mHealth Applications. In: Regional School of Machine Learning and Artificial Intelligence (ERAMIA), 2026. DOI: https://doi.org/10.5753/eramiars.2026.35112

Socials

William Niemiec © All rights reserved