Nathanael Castro

Compiling skyline0%

Software Engineer · AI Builder · South Florida

Ideas in.Products out.

I'm Nathanael Castro. I build AI products end to end, from schema to model to interface, and I ship them. The skyline behind this is generated and rendered live in your browser, because a portfolio should prove something, not just claim it.

Projects.

COPPA moderation: on parent_dash: live

Featured / React Native / iOS & Android

Truth Talks

Kids hold real conversations with AI biblical characters. The hard part was never the chat; it was building safety a parent would actually trust: layered moderation, parental controls, and COPPA compliance baked into the architecture instead of bolted on.

React NativeNode / ExpressMongoDBSupabaseOpenAI API

Read the case study

71% MACROS WEEK 3 / 12

Hackathon / Full-stack

NutriFit

An AI meal-and-workout planner built, demoed and finished inside the hackathon clock: dashboards, weekly planners, and rate-limited AI that adapts to the person instead of handing them a static PDF.

Node / ExpressSupabaseOpenAI APIEJS

Read the case study

can I afford this? No. You have $84 left and rent hits Tuesday. Skip it. F BURN RATE ↓ 38% SINCE FRANK

TypeScript / LLM product

Frank

A budgeting assistant with an actual personality. Frank is blunt on purpose. Most finance tools are passive spreadsheets, and nobody ever changed a habit because a pie chart looked concerned.

TypeScriptPrompt designPersonal finance

Read the case study

read_page() find("Sign up") click(x,y) form_input() any_llm ✓ provider-agnostic

Open source / Agents

llm‑in‑chrome

A browser agent that reads pages, decides, and acts, behind an interface that doesn't care which model you plug in. Most agents are welded to one vendor. This one isn't, and that was the whole point.

Browser automationAgent loopOpen source

Read the case study

7 STOPS · PACED FOR YOU

AI planning / Routing

El Camino

A trip planner that designs the whole journey (route, stops, and pacing) around how you actually like to travel, instead of dumping the top ten attractions and calling it an itinerary.

AI planningRoutingFull-stack

Read the case study

AI is a material, not a magic trick. It still has to be engineered.

01

Find the real constraint

Every project has one thing that decides whether it's viable. For a kids' AI app it was safety, not the model. I go looking for that first, because everything else is negotiable and it isn't.

02

Design the data before the demo

Schemas outlive prompts. I redesigned a production MongoDB layer at OneHope before touching features, because a fast demo on a bad schema is a slow product six months later.

03

Put it in front of people

I ran A/B tests and focus groups with 35 participants rather than guessing. Opinions are cheap; watching someone get confused by your UI is worth a week of debate.

04

Ship it, then watch it

Deployed isn't done. Rate limits, CloudWatch alarms, moderation logs: the boring layer is what separates a demo from something you can leave running.

Portrait of Nathanael Castro

<engineer>ai builder

Both halves are true, and that's the whole point.

I'm a CS senior at Florida Atlantic University, minoring in Artificial Intelligence & Cybersecurity, graduating May 2026. Last summer I was the sole engineering intern at OneHope, with no team to hide behind, and I shipped an AI assistant into production, redesigned the database under it, and ran the research that decided what got built.

Away from work I keep building: an app where kids talk with AI biblical characters, a budgeting assistant too blunt to be polite, an open-source browser agent. The pattern is always the same: I get curious about something, then I get stubborn about finishing it.

Based
South Florida. Open to remote & relocation
Looking for
New-grad software / AI engineering, May 2026
Currently
Building agent tooling and shipping Truth Talks
Off-hours
Hackathons, WebGL rabbit holes, and this website

A small scripted assistant.

castro.bot

Track record.

Jun to Aug 2025 · Pompano Beach, FL

AI Software Engineering Intern

OneHope

  • Sole intern engineer. Built and deployed an AI chatbot (OpenAI API, Node.js, MongoDB) that lets employees retrieve info, answer event questions, and auto-generate meeting summaries.
  • Architected a full MongoDB redesign for a production system monitored with AWS CloudWatch, improving query performance and scalability.
  • Drove feature decisions through A/B testing and focus groups with 35 participants; built the backend survey flows and shipped the frontend conversation-starter UI.

Graduating May 2026

B.S. Computer Science

Florida Atlantic University · minor in AI & Cybersecurity

  • Data structures & algorithms, database systems, software engineering, full-stack and iOS development, artificial intelligence.

Certifications

Google AI Essentials

HackerRank Python · Microsoft Foundational C# (in progress)

Languages

PythonJavaScriptTypeScriptC++JavaHTML/CSS

AI & ML

OpenAI APILLM integrationRAGEmbeddingsPrompt engineeringContent moderation

Frameworks

ReactReact NativeNode.jsExpressEJS

Data

MongoDBSupabasePostgreSQLSchema designREST APIs

Cloud & practices

AWS CloudWatchVercelGitCI/CDAgileA/B testingRate limitingCOPPA