Portfolio

Organizations adopting AI often reach a plateau where a few individuals experiment successfully while the organization itself acquires no repeatable practice. In collaboration with the Gates Foundation, we guided five professional-learning organizations through a program combining hands-on AI instruction, work-embedded lessons, and sustained coaching. Each organization completed the program adopting AI tools and methods its own staff could operate and extend.

AI models perform well in English but struggle significantly in many native African languages, limiting their usefulness for millions of disadvantaged people who stand to benefit from AI the most. In collaboration with the Bill & Melinda Gates Foundation (now the Gates Foundation), we created tools to measure the performance gap between English and African language LLMs on a variety of tasks; we used what we learned to retrain existing LLMs, and close the gap.

Electronic health records can only be searched for conditions that have already been named, characterised, and assigned a diagnosis code; for emerging conditions, establishing that definition takes years, leaving incidence unmeasurable in the meantime. In collaboration with Pfizer, we developed a three-step process to derive a condition definition from patient records alone; we applied it to post-COVID conditions, where it identified 6.6 times as many cases as the official diagnosis code.

Expert judgment becomes a bottleneck as a business grows, because every new case needs someone to weigh the evidence and explain a conclusion. In collaboration with Traive Finance, we captured how credit specialists connect agricultural and financial factors and used those relationships to drive both the risk assessment and a written report; credit analysts preferred the generated reports over those generated by their peers in up to 90% of cases.

Organizations accumulate knowledge across documents faster than people can connect it. In collaboration with the National Institutes of Health, we created BRAINWORKS to extract relationships from scientific literature, connect them to funding and authorship, and make the resulting knowledge graph available for exploration and analysis.

Companies that grow by acquisition accumulate customer knowledge in fragments, leaving analysts to reconstruct account histories by hand and data scientists to rebuild the same joins for every project. For a global software company, we replaced that reconstruction with a versioned feature store, peer-review gates, and a containerized development workflow.