At Outlier AI, I lead expert evaluation and RLHF workflows for frontier LLMs across English and Swahili. I designed evaluation rubrics that reduced hallucination rates by 28% across 50,000+ prompts.
At Handshake AI, I audited instruction-following datasets and partnered with machine learning engineers on DPO pipelines, improving cross-lingual reasoning scores by 18 points on internal benchmarks. I also authored style guides and semantic taxonomies adopted by more than 100 contributors.
Through Toptal and independent platforms, I built Swahili-English sentiment analysis and intent detection models for an East African fintech, increasing automated support resolution by 42%. My work also includes NLP pipelines for regional news feeds and a localized Swahili speech-to-text baseline co-developed at @iLabAfrica Innovation Lab.
