AI Routing Gateway
AI Routing Gateway is intended to send each query to the model or specialist workflow most likely to return a reliable answer, while keeping the routing decision explainable and testable across multiple LLMs.
Work
I work across open-source systems, private research workspaces, and early scientific concepts. The common thread is turning a difficult question into something that can be specified, tested, or made legible to the people who need to assess it.
Projects here range from prototypes and research writing to software, tools, and early ideas for reimagining existing systems. My main field is healthcare, but I also enjoy technical problems involving cars, air pollution, astronomy, life sciences, and anything that rewards a new way of thinking.
Projects
AI Routing Gateway is intended to send each query to the model or specialist workflow most likely to return a reliable answer, while keeping the routing decision explainable and testable across multiple LLMs.
AI Firewall is intended to inspect outbound data before a cloud AI call and allow, redact, or block sensitive information, giving regulated teams a visible privacy control for each request.
SystemBench is intended to evaluate an AI system across routing, tool correctness, provenance, safety, latency, operability, and task success so teams can see where the complete workflow breaks.
Smart Glasses Clinical Platform is intended to simulate clinical smart-glasses workflows and test capture, connection, voice, device contracts, and workflow states before a real deployment.
Clinical LiDAR Framework is intended to create calibrated three-dimensional wound-surface measurements with quality gates that can be compared across visits instead of relying only on similar-looking images.
Reflection AI is intended to provide consent-aware AI memory with provenance, inspection, deletion, rollback, and gated learning so people can control what personalization carries forward.
Web of Thoughts is intended to represent reasoning as typed, inspectable relations with peer review and evaluation records instead of hiding the entire path inside one answer.
Periodic Table Research is intended to test alternative layouts for representing relationships among elements while preserving standard names, symbols, and atomic numbers.
Master LLM is intended to coordinate model routing, collaboration, evaluation, privacy controls, and product infrastructure for AI systems designed for regulated fields. The implementation is private.
Clinical Feedback System is intended to collect and reward honest patient feedback, route themes to the right service owner, and support patient engagement and retention without absorbing clinical escalation. The prototype uses synthetic scenarios.
BioVOC is intended to investigate whether biological volatile organic compounds found in breath can signal cancer-related activity in the body, using a non-invasive breath sample and a traceable validation pathway.
Telomere Research is intended to study healthy aging by combining telomere biology with metabolic, inflammatory, genomic, and functional measures collected over time.
Disfigurement Index is intended to support structured, calibrated, longitudinal documentation of appearance and disfigurement without reducing a person’s experience to a single score.
Selected Clinical research
These clinical research questions show the kinds of problems I want to carry from patient experience to evidence, and eventually toward a useful system.
From question to evidence
The through-line is a difficult question moving from clinical context to study design, prototype, or system evaluation. Publication can be one step in that path, helping the work travel toward the people who need the solution.