CV · Mugume Twinamatsiko Atwine

Summary

I work on language technology and data infrastructure for health in Sub-Saharan Africa. I build and evaluate bilingual English–Luganda assistants that answer only from expert-approved or official sources, and open tools that use language models to help research projects harmonise their data.

Experience

2021 – present

Data Science Specialist, eLwazi Open Data Science Platform

Uganda Virus Research Institute, Entebbe · NIH-funded DS-I Africa program

Leads the eLwazi Data Models, Storage, Registries and Catalogue Working Group, which has had between about 20 and 50 members from DS-I Africa projects.
Co-developer of the Metadata Harmonisation Tool, an open-source application that uses large language models to map study variables to a target codebook.
Co-author (ideas and editing) of papers on the institute's bioinformatics computing service (Bioinformatics, 2026; PLOS Computational Biology, 2025).

2019 – present

Senior Data Scientist and ML Technical Lead

Africa Center of Excellence in Bioinformatics and Data-Intensive Sciences (ACE Uganda), Infectious Diseases Institute, Makerere University, Kampala

HEAL (2023 – present): team lead for a bilingual assistant that grounds GPT-4-turbo in Uganda's disease-surveillance guidelines, with a fine-tuned English–Luganda translation model. Supported by the Gates Foundation and Code for Africa. First author of the preprint.
Lacuna Luganda Health Corpus (funded by the Lacuna Fund): first author of the dataset paper, under review, describing 38,644 human-reviewed English–Luganda health sentence pairs released on Hugging Face.
AfriNorm (2026): lead developer of an open toolkit for restoring diacritics in African-language text, evaluated on six languages. First author of the paper draft.
Started two studies on Luganda machine translation (2026): how open models cope with speech that mixes Luganda and English (project lead), and predicting which machine translations need a human reviewer.
Built a model to estimate HIV risk among young people from behavioural data (SIDA Risk AI).
Developed a time-to-event model for the onset of cryptococcal meningitis in people living with HIV.
Built a retrieval-augmented question-answering system over long technical manuals.

2020 – present

Capacity Building Lead and Research Assistant Lecturer

ACE Uganda, Makerere University, Kampala

Developed and taught a Big Data Analytics course for MSc and PhD students.
Supervised student research projects.

2021 – 2023

Natural Language Specialist (contract)

Academy for Health Innovation, Infectious Diseases Institute, Kampala

Led the technical build of a clinical decision support system for health facilities.

2021 – 2023

Lead Data Scientist, COVID-19 response

Makerere University, Kampala

Technical lead for COAST, an English–Luganda COVID-19 chatbot that answers only with expert-approved responses. Built with Rasa and deployed on Kubernetes. Co-first author of the resulting paper in PLOS Digital Health (2026).

Education

2011 – 2013

MSc Computer Networking

Limkokwing University of Creative Technology, Malaysia

Research on workload optimisation and load balancing in virtualised cloud data centres, published as four papers (listed below).

2009 – 2011

BSc Computer Science

Mbarara University of Science and Technology, Uganda

Certificates

Computational Thinking and Big Data, University of Adelaide (edX), 2017
Project Management, University of Adelaide (edX), 2018
Evidence-Based Project Management, Australian National University (edX), 2018

Publications

Journal articles

Atwine MT*, Jjingo D*, Nsubuga M, Serunjogi R, Mbabaali I, Lujumba I, Byansi D, Kintu T, Galiwango R, Kebirungi G. A bi-lingual chatbot implementation for pandemic response using the transformer-based approach. PLOS Digital Health. 2026;5(4):e0001256. doi:10.1371/journal.pdig.0001256. *Co-first authors.
Lukyamuzi E, Kimbowa Wamala T, Ssekagiri A, Galiwango R, Kebirungi G, Atwine M, Nsubuga M, Maslamoney S, Panji S, Mulder NJ, Jjingo D, Kayondo J. A scalable HPC framework for bioinformatics in resource-limited settings: design principles, implementation, and sustainability from the UVRI experience. Bioinformatics. 2026;42(4):btag149. doi:10.1093/bioinformatics/btag149
Jjingo D, Walakira A, Hashim S, et al. (Atwine MT, 36th of 67 authors). Pathways, outputs and impact of NIH-supported bioinformatics and genomics graduate trainees in Africa. Briefings in Bioinformatics. 2026;27(1):bbag046. doi:10.1093/bib/bbag046
Galiwango R, Whalen CJ, Kebirungi G, Atwine MT, et al. (26 authors). Ten simple rules for building and maintaining sustainable high-performance computing infrastructure for research in resource-limited settings. PLOS Computational Biology. 2025;21(9):e1013481. doi:10.1371/journal.pcbi.1013481

Preprints, papers under review and in preparation

Atwine MT, Kintu TM, Lujumba I, Mbabali I, Muwonge K, Muwonge L, Nakiire L, Ntono V, Nsubuga M, Galiwango RM, Lunkuse S, Kebirungi G, Jjingo D. Development and evaluation of HEAL: a bilingual English–Luganda RAG chatbot for disease surveillance in Uganda. Research Square [preprint]. 2026. doi:10.21203/rs.3.rs-10357473/v1. Under review.
Atwine MT, Kintu TM, Kebirungi G, Muwonge L, Muwonge K, Lujumba I, Mutegeki H, Kyagambiddwa T, Jjingo D. A human-reviewed English–Luganda parallel corpus for natural language application in biomedical communication. Under review. Dataset: doi:10.57967/hf/10682
Atwine MT, Muwonge L, Kintu T, Muwonge K, Galiwango R, Kebirungi G, Jjingo D. AfriNorm: an open, extensible system for diacritic restoration in African languages. In preparation.
Adams R, Adeleke F, Junck L, et al. (Atwine MT, 52nd of 75 authors). Mapping the potentials and limitations of using generative AI technologies to address socio-economic challenges in LMICs [version 1]. VeriXiv [preprint]. 2025. doi:10.12688/verixiv.948.1

Earlier publications (MSc research, published as Twinamatsiko AM)

Twinamatsiko AM, Abdulhussein ANA, Joshi JH, Lashkari AH, Sadeghi M. Dynamic workload performance optimization model for multiple-tenancy in cloud systems. International Journal of Computer Science Issues. 2013;10(6):271.
Parhizkar B, Abdulhussein ANA, Joshi JH, Twinamatsiko AM. A common factors analysis on cloud computing models. International Journal of Computer Science Issues. 2013;10(2 Pt 1):523.
Joshi JH, Abdulhussein ANA, Twinamatsiko AM, Lashkari AH, Sadeghi M. An enhanced quality model for IaaS provider with non-quantifiable key performance indicators. International Journal of Computer and Communication Engineering. 2014;3(2).
Abdulhussein ANA, Joshi JH, Twinamatsiko AM, Lashkari AH, Sadeghi M. An efficient load balancing algorithm for virtualized cloud data centers. In: Recent Advances in Electrical and Computer Engineering (conference proceedings).

Software and data

Lacuna Luganda Health Corpus. 38,644 English–Luganda health sentence pairs with a review history for every pair and fixed training, development and test splits. Hugging Face, 2026. CC BY 4.0. doi:10.57967/hf/10682
Metadata Harmonisation Tool, version 0.4.3. Open-source application that uses large language models to map study variables to a target codebook. Zenodo, 2026. doi:10.5281/zenodo.18978529; code at github.com/atwine/Metadata-Harmonisation-Tool
COAST chatbot dataset. Expert-approved COVID-19 questions and answers in English and Luganda. Zenodo, 2025. doi:10.5281/zenodo.14608441

Technical skills

Language models and NLP: retrieval-augmented generation, the GPT-4 API, fine-tuning translation models (mBART) and byte-level models (ByT5), machine translation evaluation, Rasa
Machine learning: classification, time-to-event (survival) models, model evaluation with domain experts
Engineering: Python, PyTorch, Hugging Face Transformers, Streamlit, Flask, React, Docker, Kubernetes (K3s)
Data: data harmonisation, FAIR metadata, research data pipelines

References

Available on request.

Last updated November 2026. Online version: atwine.github.io/cv