Hi, I'm
> Software engineer
I build models that answer real questions, and the software around them. Creator of Once Upon YOUR Child, NSF REU researcher, and author of a research abstract accepted to the 2026 ACM SIGSPATIAL Student Research Competition (undergraduate track).
About me
What's really going on, and what would actually change it? That's the question behind everything I build.
Two things right now. I'm an undergraduate research fellow on Portland State's Computational Modeling REU, where I built an agent-based model of Portland's street network that produces per-segment NO2 emission and noise estimates. Close one arterial in the model and you can see exactly which parallel streets absorb the traffic and the pollution, before the lane closes rather than after.
The other is Once Upon YOUR Child, an AI storytelling app for kids that I designed and built under my own LLC: Flutter on the front, Flask, Celery, and Redis on the back, a parental-consent flow designed around COPPA. One feature explains the rest: I built a limerick mode because I know an eight-year-old who can't read and refuses anything that looks like a baby book. She'll read something she's the hero of.
Experience
Aug 2026 – Oct 2026
Continuing the REU project · with Dr. Christof Teuscher, Portland State · remote
Banked the model's numeric predictions for the I-5 Rose Quarter closure in a public preregistration before the closure began. A public travel-time log recorded what happened on 12 frozen routes, so the predictions can be graded against reality. Results are held for the paper.
Jun 2026 – Aug 2026
Portland State University · Computational Modeling Serving Portland · remote
Built an agent-based traffic simulation that produces street-level NO2 emission and noise estimates on a 159,000-segment network, checked against observed traffic counts on 356 matched segments (rank correlation 0.59). Multi-seed experiments on the university's Slurm cluster with pinned seeds and checkpoint/resume.
Nov 2025 – Present
Once Upon YOUR Child LLC
Designed and built an AI storytelling app for kids: a Flask/PostgreSQL API, Celery workers over Redis, a Flutter client, and Docker images deployed through GitHub Actions with a rollback path.
Expected Mar 2028
Colorado State University Global · GPA 3.97 / 4.0
Projects
Personalized AI storytelling app for kids' social-emotional learning, designed and built under my own LLC. Flask/PostgreSQL API with JWT auth, rate limiting, and caching; Celery workers over Redis; Flutter web client; Docker image through CI/CD with a rollback path.
Agent-based traffic simulation over Portland's street network with car-following, signal queueing, and shortest-path rerouting, producing per-segment NO2 emission and noise estimates. Checked against observed counts on 356 matched segments; multi-seed runs on a Slurm cluster. NSF REU at Portland State.
Personalized guided-meditation app. A short questionnaire about the life you want becomes a script written for you, read by the voice you choose over a layered soundscape, with singing-bowl bells placed where the script calls for them. Flutter app; Flask API with PostgreSQL and Redis-backed workers; Firebase Hosting and Railway. In private testing.
Should I run outside? A running advisor that fans out to weather, air-quality, and severe-weather APIs with structured concurrency, models results as records and sealed interfaces, and scores every hour against your own tolerances. An entry in Hackster's "Modern Java in the Wild" 2026.
iOS app for practicing difficult conversations with an AI persona that pushes back realistically and scores your approach. A Hono API on Cloudflare Workers isolates every secret and enforces rate limits.
A Model Context Protocol server that lets Alexa+ tell personalized, feelings-aware bedtime stories from Once Upon YOUR Child. Exposes the story engine to an LLM agent as typed tools.
Java 26 CLI and web tool that healthifies any recipe with explainable ingredient swaps and a local recipe library. Zero runtime dependencies, a JUnit 5 suite, and GitHub Actions CI.
More on GitHub
Research
61–76%average drop in estimated NO2 emissions on a closed arterial, across three arterials and 12 random seeds
12/12seeds where estimated NO2 emissions rose on both parallel detours when SE Powell closed
The traffic doesn't disappear when a street closes. It reroutes, and the pollution goes with it. In the model, SE Division and SE Holgate picked up about 32 and 42 percent more estimated NO2 emissions on average. That's the kind of thing a transportation planner wants to know before the lane closes rather than after.
A static land-use model cannot show this: its inputs do not change when a road closes, so it predicts zero change on every segment. The comparison is model against model, because Portland has no dense sensor network to check either one against.
Publications
Talks & open science
Skills
Production backends, typed front ends, and research code that has to be reproducible. I care about tests that run in CI, deploys that can be rolled back, and numbers I can stand behind.
Computational neuroscience and EEG signal processing, where modeling and signal work meet a problem I care about.
Languages
Web & backend
Tooling & delivery
CS foundations
Every change starts as a tracked ticket, goes through a pull request, and deploys with a rollback path.
Tracing, health endpoints, queue-backlog alerts, backups to object storage, and an automated restore drill that checks row counts.
A market-validation study, a rebrand after finding naming collisions, and copy rewritten to clear state statutes and App Store guidelines.
A SIGSPATIAL research abstract, an REU proceedings chapter, Ignite talk at the PSU research symposium, and READMEs that explain the one rule the structure enforces.
Modeling & simulation
Data & ML
Compute
Contact
Happy to talk about research collaborations, internships, and interesting problems. Send a message below, or email me directly at darcy0408@gmail.com.
