Portfolio — Research Operations, External Artifacts

This post describes an AI-driven data visualization system built for laboratory environments, combining AWS services with machine learning pipelines. It demonstrates my comfort editing AI/ML subject matter and producing clear prose from technical authors whose primary expertise is not writing.

This post makes a technical argument for an AWS-based approach to remote clinical trial monitoring. The editing challenge was helping the author present a clear, evidence-grounded case for their approach while acknowledging its scope and limitations honestly—a challenge familiar to anyone editing safety or evaluation documentation.

A technically dense post on security hardening for Kubernetes clusters. This sample demonstrates that I can work fluently with highly specialized infrastructure content, preserving precision while improving structure and readability for a knowledgeable audience. The editing task was largely structural: reorganizing the argument so that the author’s reasoning was visible, not just their conclusions.

A data engineering post aimed at practitioners migrating large-scale data infrastructure, this sample shows range across the data domain and illustrates my ability to edit content I did not fully understand at first read—asking the right questions, checking terminology against source documentation, and producing prose that is accurate because I did not guess.

This 158-page guide was written and compiled for the Technical Communications team at NCBRT, where I served as Adjunct Editor for nearly eleven years. It covers abbreviations, capitalization, citations, grammar, numbers, punctuation, source formatting, and usage specific to emergency management and federal training contexts.

Building it required auditing inconsistencies across years of existing course materials, making judgment calls about contested usage, and writing guidance clearly enough that non-editor subject-matter experts could apply it independently. This is the kind of work Anthropic describes as developing templates, style guidance, and contributor guides so that each production cycle starts from a stronger baseline.

This 32-page guide was written for writers and editors working on the prekindergarten through fifth-grade Eureka Math curriculum—a large, multi-author document production effort with tight consistency requirements across hundreds of lessons and problem sets. Based on The Chicago Manual of Style, it extends that base to address the specific demands of mathematics curriculum: notation conventions, number treatment in mathematical prose, capitalization of curriculum-specific terms, typesetting of problem types, and a glossary of preferred forms for language that recurred throughout the project.

The editorial challenge was precision at scale: many authors writing to the same standard, in a domain where a misplaced hyphen or an inconsistently named problem type produces real errors for teachers and students. Writing guidance that non-editor writers could follow without editorial supervision was the primary design constraint.

This 38-page plan sets FEMA’s agency-wide goals and objectives for a five-year period across three strategic priorities: equity, climate resilience, and agency readiness. It is a public commitment document: FEMA issued it knowing it would be used to evaluate whether the agency delivered on its stated goals.

The editorial work required maintaining a consistent institutional voice across contributions from multiple internal stakeholders, ensuring that the policy language was precise rather than aspirational, and catching commitments that were phrased in ways that couldn’t be evaluated or held to account. A strategic plan that cannot be measured against is not a plan—it is a press release. The distinction between the two is an editorial judgment as much as a policy one.

This is the most direct parallel in my portfolio to system card production. Both are documents in which an organization goes on record about what it does and does not claim, and accepts public accountability for those claims.

This regional emergency operations plan was produced for FEMA Region 10 in preparation for a Cascadia Subduction Zone seismic event—one of the highest-consequence natural disaster scenarios in North America. The document is multi-agency, multi-author, and structured to be used operationally under crisis conditions, where clarity and precision are not abstract virtues but practical requirements.

The editing challenges were structural as much as stylistic: ensuring consistent terminology across authors who wrote independently, maintaining logical flow across sections with different owners, verifying that cross-references and role assignments were internally coherent, and catching claims that subtly contradicted each other across sections. The document had to read as one coherent plan, not a collection of contributions.

At 2,557 pages covering more than twenty participating jurisdictions, this plan represents the largest class of document in my regular editorial practice. Hazard mitigation plans of this scale are produced by assembling dozens of individually authored Word files—written by planners, subject-matter experts, and jurisdiction representatives across the county—into a single unified document, which is then edited for consistency, coherence, and compliance with FEMA formatting and content standards before submission.

The editorial process is genuinely collaborative: the IEM Technical Communications Team contributes at multiple stages, with different editors taking ownership of different sections and phases. On this project, an outside editing firm was also engaged to manage volume; IEM editors, including me, reviewed and reconciled their work alongside our own. Final quality review of the combined document—checking for terminology drift, broken cross-references, inconsistent role assignments, and formatting compliance—requires one editor to hold the whole in view at once.

The document goes through multiple submission cycles: to the client for review, back to IEM for revisions, to the state for approval, back again, and ultimately to FEMA. FEMA approval is required for jurisdictions to remain eligible for federal hazard mitigation planning grants—which means the accuracy and internal consistency of the document is not an editorial nicety but a condition of federal funding.

I am currently co-authoring a structured evaluation of AI writing and editing tools for IEM’s Technical Communications Department, where I am leading the assessment. The tools under evaluation are Grammarly, Microsoft Copilot, Claude, and ChatGPT, assessed against IEM’s actual editorial workflow: tracked-changes-based document review, multi-thousand-page deliverables, Section 508 compliance, and FEMA submission requirements.

The completed sections evaluate Grammarly and Microsoft Copilot in detail. The Grammarly analysis identifies a fatal workflow incompatibility—accepting a suggestion replaces the entire paragraph in Track Changes rather than marking only the changed text, destroying the audit trail IEM’s review process depends on—and distinguishes between what the tool is and is not suited for in a professional editorial context. The Copilot analysis documents a pattern of false confidence: the tool consistently presented the appearance of completing tasks it had silently failed to complete, which the report characterizes as a transparency problem rather than merely a capability limitation. A second finding is equally relevant to professional editorial work: when Copilot did produce output, it made extensive unsolicited changes—rewriting, restructuring, and deleting content it was never asked to touch, silently and without consultation.

This document is an internal work product and is not posted here publicly. I am glad to share it directly with Anthropic on request.