Skip to content

Bryan Beltrán

Software Engineer II at Chewy — event-driven systems, observability, and developer tooling

About

I work on Chewy's Agent Experience team, building systems for customer care across chat, phone, and email. I was recently admitted to Georgia Tech's Master of Science in Computer Science program and will begin in Spring 2027, following the Machine Learning track. Side projects solve problems in my own life and give me room to learn.

More about me →

Selected Work

Flagship Project

SeedStarter

Launched

Production-style frost-aware garden planner with climate-backed schedules and exports.

  • 88 crops and 2,000 varieties with season-aware planting timelines.
  • NOAA climate percentiles and fallback resolution for roughly 33,000 US ZIP codes.
  • OpenAPI docs, golden climate evaluations, ADRs, and production smoke tests.

Room for Words

Launched

Calm, local-first literacy companion for K–2 learners practicing decoding and phonics fluency.

  • Brief practice rounds for regular-word decoding, grapheme–phoneme patterns, and comfortable fluency.
  • Teacher catalog, adult-managed learner tools, and privacy-preserving local practice boundaries.
  • Buildless web app with accessible interactions and a D1-backed worker API for optional sync.

Voiceprint

In development

Local-first personal voice compiler that turns interviews and writing samples into inspectable, versioned instructions for AI tools.

  • Guided interview and corpus modes produce an inspectable canonical writing profile.
  • Deterministic adapters export portable instructions for Codex, Claude, ChatGPT, Cursor, and Markdown.
  • Local persistence, version history, diffing, and blind A/B evaluation keep the workflow user-owned.

Math Garden

In development

Production-oriented, accessible K–5 math practice platform with teacher-authored game versions and anonymous learner play.

  • Eight playable templates with deterministic sessions, feedback, hints, and refresh recovery.
  • Teacher review, publication, immutable versions, rollback, assignments, and a curated library.
  • Minnesota K–5 standards coverage with explicit accessibility, privacy, and content-review boundaries.

Room for Math

In development

Buildless teacher-facing library of ready-to-use K–5 math games with local custom-game creation.

  • 57 preloaded games spanning Kindergarten through Grade 5 and Minnesota’s 2022 math strands.
  • Search and filters with teacher guidance, detail views, and short playable previews.
  • Device-local custom-game builder with a responsive, dependency-free browser runtime.

Browser Listener

In development

Local-first MV3 Chrome extension for recording authorized browser sessions into portable, redacted bug-reproduction and evidence bundles. No upload or telemetry.

Wiggum

In development

Auditable Python harness for autonomous software iteration with isolated worktrees, validation gates, review, and evidence.

GovDataHub

In development

Private prototype for a commercially reusable Census data service with catalog discovery, normalized queries, caching, and provenance.

  • FastAPI facade over ACS, business, and other U.S. Census datasets.
  • Variable discovery, normalized queries, saved requests, and CSV/JSON export.
  • Source URLs, fetch times, cache state, and checksums travel with every result.
See all projects →

From the Blog

Notes on tools I'm building and exploring.

Building a Privacy-First Debug Capture Extension on Manifest V3

MV3 capture architecture from production customer-care tooling — CDP, redaction, service worker constraints, and why nothing leaves the browser.

Visit Blog →