CSU-Global B.S. CS (3.97 GPA) • Undergraduate research fellow • Low-compute, reproducible AI • Flutter & TypeScript apps I build and operate end to end

• Ship and operate, not just prototype: production deploys, automated rollback, tested database recovery.
• Reproducibility: seeds, exact params, and restart-safe logs baked into every research project.
• Accessible AI: CPU-first pipelines that run on modest hardware.

Languages: Python, TypeScript/JavaScript, Dart, Java, SQL Frameworks: Flutter, React Native (Expo), Flask, FastAPI Infra & Dev: Docker, PostgreSQL, GitHub Actions (CI/CD), Cloudflare Workers, Railway, REST APIs Research: Agent-based modeling, OSMnx, model validation, spatial data

An agent-based model of interacting vehicles on Portland's street network. Vehicles follow one another, queue at signals, and back up in congestion; from those interactions the model produces street-segment surfaces of traffic NO2 and noise.

• The agent simulation generates the predictors, which are then fed into the same random-forest method a published land-use baseline used — so the comparison isolates what source-based interaction modeling adds over static estimation
• Closure experiments quantify that 61–76% of a closed arterial's emissions redistribute onto specific parallel routes, naming which blocks inherit the burden
• HBEFA emission factors, CNOSSOS noise modeling benchmarked against the FHWA Traffic Noise Model
• Abstract accepted to the ACM SIGSPATIAL 2026 Student Research Competition (undergraduate track)
• Repo: portland-traffic-abm

A production cross-platform app generating personalized, age-appropriate stories that help kids build feelings vocabulary and social-emotional skills. Built and operated solo.

• Flutter front end · Python/Flask API · PostgreSQL · Docker on Railway
• 10 CI/CD pipelines — canary deployment, automated rollback, scheduled Postgres backups, and restore drills that verify backups actually recover
• 2,800+ commits and 220 test files over nine months of continuous delivery
• Live: onceuponyourchild.app · Repo: once-upon-your-child

A Python model of how a false rumor spreads through a population in Portland — exploring when misinformation settles, oscillates, or can be reversed by debunking.

• Built for the PSU REU, “Computational Modeling Serving Portland”
• Repo: misinformation-dynamics-model

An agent-based model asking whether a town fleeing a wildfire is safer leaving all at once or in staggered waves. Traffic jams and smoke exposure emerge from agent interaction rather than being scripted.

• Finds a density tipping point where the optimal strategy flips — and shows it disappears when the fire spreads fast
• Repo: wildfire-evacuation-abm

An iOS app for practicing difficult conversations with an AI persona that pushes back realistically, then scores clarity, composure, and assertiveness.

• Cloudflare Worker proxies all model and TTS calls, so no API credentials ship in the client; speech-to-text runs on-device
• 73 Worker tests passing; TypeScript end to end (Expo / React Native + Cloudflare Workers)
• Parked as a working prototype and published as a reference implementation — Repo: soundingboard

Interactive satellite screensaver touring Earth’s wonders, as a web app and a Windows .scr.

• Live weather overlays and a multi-year imagery time-slider
• Repo: drift-screensaver

• Building out Story Weaver toward store release — COPPA-aware consent flow, age-banded content, and release automation.
• Writing up the REU modeling work; abstract accepted to the ACM SIGSPATIAL 2026 Student Research Competition (undergraduate track).

• Email: darcy0408@gmail.com
• LinkedIn: https://www.linkedin.com/in/darcyvanpelt

_{Always happy to chat about accessible AI, reproducibility, and kid-centered apps.}