I am a data science leader specializing in the intersection of machine learning and infectious disease. By combining a Master of Public Health and Epidemiology with a Master of Science in Data Analytics, I build targeted artificial intelligence systems designed for the realities of field medicine.
Currently, I am developing Arclight-FEBE, a proof-of-concept small language model engineered for the epidemiology domain. It serves as the foundation for the broader Arclight reference architecture: a framework for deploying local, edge-capable language models that prioritize data privacy, offline functionality, and operational resilience in restricted environments.
The Arclight Thesis
The future of AI belongs to local, private, offline models running on edge devices which are capable of accomplishing 95% of real-world use cases.
Local AI Follows the Arc of History
We’ve seen this path before, from the earliest days of digital devices. Computer mainframes were initially tested and experimented with in academic and commercial labs. Computation then spread to enterprise and government end users via centralized mainframes leased or sold to enterprises, with proprietary restrictions. Continued developments and improvements led to personal computers, and, years later, smartphones, following consumer demand for convenience, personalization, privacy, and sovereignty. As devices became more accessible and personal, demand exploded, and led to follow-on technologies like the internet, social media, and an entirely new on-demand digitally enabled economic sector, all of which would be impossible without the move from mainframe to smartphone.
AI Needs a Sustainable Path Forward
Cutting-edge foundation models running on massive data centers are financially and ecologically unsustainable, and unnecessary for 95% of real-world end uses. Custom-built local AI implementations, tuned for specific end uses, and running on edge devices, are environmentally friendly, require no incremental inference costs, isolate user data by design, and remove vulnerabilities to token cost inflation or changes to terms and services agreements. This design philosophy could increase access and adoption globally, with countless follow-on benefits similar to those seen with the democratization of computation through smartphone proliferation.
Arclight is a reference architecture designed to make this dream possible.
Read the Essay: The Frontier Fallacy.