Research

I develop machine learning systems that tackle real-world challenges in low-resource and high-impact settings. As a Senior Applied Research Scientist at Microsoft AI for Good Lab, my work spans geospatial AI, multilingual NLP, and interpretable machine learning. I focus on building scalable, inclusive, and transparent AI tools that support climate resilience, disaster response, and equitable access to information.

🌍 Geospatial AI & Remote Sensing

I develop machine learning models for analyzing satellite and aerial imagery to support environmental monitoring, agriculture, and disaster response. My research projects include:

• GeoVision Labeler (2025): A zero-shot geospatial classification framework using vision-language models to label satellite imagery without manual annotations.

• Expanding Smallholder Irrigation in Central Kenya (2025): Published in Environmental Research Letters, this study maps the rapid expansion of irrigated agriculture in Laikipia County, Kenya, and highlights the environmental risks posed to grassland ecosystems and wildlife due to water extraction and land conversion.

• Weak Labeling for Cropland Mapping in Africa (2024): Presented at the IEEE International Geoscience and Remote Sensing Symposium (IGARSS 2024), this paper introduces a scalable approach to cropland detection using weak supervision. It refines global cropland maps with unsupervised object clustering and sparse human annotations to train a semantic segmentation model, achieving significant improvements in F1 score with minimal labeled data.

• Analyzing Environmental Change in Namibia (2024): Presented at STAI’24: International Workshop on Sustainable Transition with AI, collocated with the 33rd International Joint Conference on Artificial Intelligence (IJCAI 2024) in Jeju, South Korea. This study leverages archival aerial photography and deep learning to map land-use changes in Namibia over several decades, highlighting the value of historical imagery in understanding long-term environmental transformations.

🌐 Multilingual NLP for African Languages

I contribute to the development of inclusive NLP tools for underrepresented languages:

• AfriQA (2023): A cross-lingual open-retrieval question answering dataset covering 10 African languages.

• English2Gbe (2021): A multilingual machine translation model for Fon and Ewe, presented at NeurIPS ML4D.

• FonMTL (2023): A multitask learning framework for the Fon language, advancing low-resource NLP.

🧠 Interpretable & Econometric Machine Learning

I explore the intersection of classical econometrics and modern ML to build interpretable models for policy and economic analysis:

• GAM(L)A (2022): The arXiv preprint version of this work, introducing a hybrid model that combines Generalized Additive Models with Lasso regularization for interpretable predictions in economic data.

• Interpretable Machine Learning Using Partial Linear Models (2024): The peer-reviewed and extended version of GAM(L)A, published in the Oxford Bulletin of Economics and Statistics.

For a full list of my publications, visit my Google Scholar profile.