Portfolio
An Adversarial, Multi-Round Novelty Audit
A seven-round literature audit across the open web, arXiv, and library databases -- over eighty papers read in full, every claim source-tagged, and the research contribution repeatedly narrowed and re-based as closer prior work surfaced.
Read case study →
Designing a Decision-Safety Mechanism for Autonomous Agents
Design-science research specifying a deterministic act/defer/abstain safety check for whether an LLM agent's evidence actually supports its next action -- architecture and formal properties established, empirical evaluation pre-registered but not yet run.
Read case study →
A Reproducible Research-Cataloguing Workflow
Designed a project-agnostic cataloguing standard -- content hashing, stable identifiers, explicit classification -- for a 90+ source literature base, so the archive is reproducible and resilient to lost files or a failed database.
Read case study →
A Zotero Write-and-Mirror Pipeline, and a Lossless Database Recovery
Extended a Python tool against the Zotero Web API to mirror 97 sources with attached PDFs, fixed two real defects found under load, and recovered an apparently corrupted local database without loss by diagnosing an OS-level cloud-storage eviction instead of rebuilding from scratch.
Read case study →