Building Clinical AI, Healthcare Data & Medication Safety Systems

Problem · Fine-tuning large language models on clinical data requires specialized infrastructure and expertise that most healthcare AI teams lack.

Solution · Production-ready pipeline for healthcare language model training using Google Cloud TPU systems, enabling efficient fine-tuning of clinical LLMs with PEFT techniques including LoRA and FSDPv2.

Reduces the infrastructure barrier for healthcare AI research teams attempting to fine-tune domain-specific clinical language models at scale.

Problem · Drug safety surveillance datasets for African populations are severely underrepresented in global pharmacovigilance databases, limiting the ability to identify adverse drug reactions in these populations.

Solution · Open pharmacovigilance dataset focused on adverse drug reactions across Africa, designed for use in drug safety research, healthcare AI training, and clinical pharmacology studies.

Published dataset supporting pharmacovigilance research and healthcare AI development for underrepresented African patient populations.

Problem · Early identification of diabetes risk requires accessible, data-driven tools that clinicians and researchers can use without complex infrastructure.

Solution · Machine learning model for diabetes risk assessment trained on clinical data, published on Kaggle for open access by researchers and healthcare AI teams.

Publicly available clinical ML model supporting diabetes risk research and healthcare AI education.