R&D projects
Active and exploratory research projects. Each one targets a specific problem at the intersection of AI, automation, and infrastructure.
Designing a coordination layer for multi-agent systems. The protocol handles task decomposition, assignment, validation, and recomposition. Current focus is on defining the handoff interface between agents and the checkpoint mechanism that ensures quality at each stage.
Exploring how to run capable AI models on-device without relying on cloud backends. Quantisation, model selection, caching, and graceful fallback when local resources are insufficient. The goal is 80% cloud independence for common workloads.
The Webman system defines a repeatable workflow for building production websites. Skills for planning, content, brand definition, creation, auditing, and deployment. Every site built with Webman follows the same auditable path from idea to live.
How do you scale AI inference without scaling costs proportionally? Research into batching strategies, model routing, and adaptive quality settings that maintain user experience while reducing compute spend.