Resume

SPAR — Research Fellow, Part-time

September 2026–Present

Conducting concept-injection experiments to investigate whether LLM self-reports reflect internal model states or are shaped only by the supplied context.

February 2026–Present

Building code-defined agent pipelines callable by other programs, wrapping Claude Code, Codex, and a custom harness with deterministic steps for worktree setup and CI coordination.
Built evaluations from recorded engineering decisions to measure agreement with my rulings; after prompt and RAG approaches plateaued at 65% agreement, began fine-tuning a model to improve decision quality.
Designing and building a hybrid RAG system for coding agents, combining graph and vector retrieval with source-version history to assemble related context in a single request.

May 2022–January 2026 · Remote

Traced undocumented enrollment logic in a legacy Python application to connect student enrollment with chat provisioning and group management through the Slack API.
Adapted a legacy Python application's authentication flow during an Auth0 migration to support machine-to-machine tokens, enabling Retool support tools to call application services.
Replaced an unreliable AI-based student scheduler with a deterministic TypeScript implementation using a min-conflicts algorithm and a schedule-conflict verifier.
Initiated and built an upgrade-testing tool that exercised database reads and writes to catch failures that connectivity checks missed during a prior P1 outage.
Interviewed support teams to identify undocumented workflows using direct database access, then replaced those workflows with permission-scoped Retool tools.
Built a support UI and Python workflow to reset lesson progress in an edge case, replacing manual deletion while preserving submission records.

December 2019–August 2021 · Denver, CO

Built Kotlin migration tooling with MyBatis to map fields between legacy and replacement database schemas and check that records imported and exported correctly.
Built Vue and JavaScript UI components that masked sensitive input during identity verification.

November 2016–July 2019 · Fort Collins, CO

Worked directly with residential customers to understand how they wanted their homes automated, adapt the implementation to their preferences, and explain technical constraints.
Traced failures during a large residential voice-control installation to an undocumented Google device limit and built a Python workaround that allowed the installation to proceed.
Evaluated software options across the home-automation stack and chose an approach combining existing devices with custom integration software.

June 2015–August 2016 · Fort Collins, CO

Built a Python server-inventory application that reduced the quarterly two-day inventory to a simple two-hour checklist.