Innovations for Poverty Action (IPA) is a research and policy nonprofit that creates and shares evidence while equipping decision-makers to use evidence to reduce poverty. IPA leads the field of development in cutting-edge research quality, innovation, and impact, making it ever more important for them to have a high-performing infrastructure to ingest, store and utilize research to aid impoverished communities. Microsoft AI for Humanitarian Action & Microsoft Philanthropies partnered with Azure Data & AI Designated Solutions Partner, Valorem Reply, to help IPA build an analysis environment that can use standard data science software with storage and computational specifications adequate for estimating a pre-coded neural network model.

CHALLENGE

• Constrained on local machines and limited in the kinds of models that IPA can run.
• Geospatial data & wealth maps include a long list of datasets.
• Geospatial datasets need to be incorporated into machine learning models to increase the level of accuracy.

SOLUTION

• Build an Azure environment with a high performing and scalable compute virtual machine with Ubuntu Linux.
• Add and load datasets to Data Factory and Azure Storage.
• Copy existing ML models to Notebooks in ML Workspace.

RESULT

• A streamlined method of importing the training data and re-calibrating the Poverty Probability Index model.
• Improved Analysis with AI/ML-transformed satellite imagery and unstructured data.
• Scalable, highly available, and performant solution to support the future needs of IPA.