Open research, permanent records
Research
I publish what I measure. Every item below has a permanent DOI, an open dataset or public source code, and can be cited. The work sits at the intersection of distributed systems and the infrastructure AI agents actually run on.
Papers
Type Coercion Gaps in Distributed SQL Engines: A Case Study of VARCHAR-to-INTERVAL Casting in Trino
Distributed query engines grow federated type systems over years, and individual coercions quietly go unimplemented. This paper documents one in Trino: the missing cast from VARCHAR to INTERVAL DAY TO SECOND, reported by a project founder in 2020 and unresolved for six years, even though the reverse direction shipped with the engine. Covers the practical cost to text-based data pipelines, the decision to reuse the existing literal parser rather than write a second one, and a correctness lesson that continuous integration surfaced during the fix.
Read the paper PreprintBiomimetic Principles in Distributed and AI Systems: A Perspective on Nature-Inspired Architectures for Resilience and Efficiency
Most bio-inspired computing is presented as a catalogue of metaphors. This paper argues the durable value is structural rather than poetic. It introduces the Biomimetic Systems Map, applies it across five well-grounded mechanism families including stigmergy, spiking computation and quorum sensing, and extracts four transferable design invariants: locality, redundancy, feedback and degeneracy. It closes with five falsifiable propositions rather than claims, and is explicit that analogy generates hypotheses, not proofs.
Read the paperSoftware and datasets
mcp-probe: a conformance and reliability checker for Model Context Protocol servers
A dependency-light command line tool that speaks the Model Context Protocol directly over stdio. It runs the initialize handshake, enumerates advertised tools, validates every tool schema against the specification, checks whether safety annotations are declared, measures latency and exercises safe round trips. Built as both a developer diagnostic and a reproducible instrument for empirical study of the MCP ecosystem. Archived with a permanent DOI; currently under review at the Journal of Open Source Software.
View the archive Source and datamcp-probe source, benchmark frame and raw measurements
The live repository, including the sampling frames and the raw per-server measurement records behind the ecosystem study. Published so the numbers can be checked rather than taken on trust: anyone can rerun the instrument against the same frame, or against their own servers, and compare.
Browse the repositoryOpen source contributions
Citing this work
Each item above resolves through a permanent DOI, so links will not rot. The canonical author record, including anything published after this page was last updated, is the ORCID profile.