DaríoCabascango

UX Engineer and AI-native developer

I build web and mobile applications with scalable architecture and great user experiences, shipped faster with AI agents.

Email me GitHub LinkedIn

Experience: 5+ years
Based in: Quito, Ecuador (UTC−5)
Languages: Spanish (native), English (B2)

About me

Hello! I’m Darío. For 5+ years I’ve built reliable applications for web and mobile.

My goal is to build products and fast experiences that people enjoy using. I care about clean architecture, accessibility and code that stays maintainable long-term.

I work AI-native: agents, skills and MCP servers are part of my daily workflow, from the spec to the code review.

Tech stack

Languages: TypeScript
JavaScript
Python
Frontend: React
React Native
Astro
Backend: Node.js
Testing: Jest
Storybook
AI tools: Claude Code
OpenCode
Codex
Figma MCP
Engram

UX engineering

I work between design and engineering: Figma becomes components that look right, feel right and work for everyone.

Design systems
Tokens and components shared by web and mobile, kept in sync with Figma.
Accessibility
WCAG 2.2 AA as the baseline: semantic HTML, keyboard support, screen reader labels and contrast checks.
Interaction & motion
Motion that explains what changed and respects reduced-motion settings.
Verified in the browser
Interaction tests and visual checks against the design before anything ships.

AI-native engineering

I build agent workflows that ship production code: spec first, specialized agents, verified output.

Agent orchestration
Planner, builder, QA and reviewer agents, each with one job and clear handoffs.
Skills & MCP
Reusable skills and MCP servers that give agents real context: the design system, Figma, Jira.
Quality gates
Claude and Codex review each other’s work, plus automated iOS and Android QA and Figma fidelity checks.

Concepts I work with

Agent core

Context & memory

Capabilities

Multi-agent

Reliability & safety

Workflow

Agent core

Agent loop

The cycle an agent repeats: read the context, decide, call a tool, observe the result, until the task is done.

Agent loop

The cycle an agent repeats: read the context, decide, call a tool, observe the result, until the task is done.

Harness

The runtime around the model: tools, permissions, context management and the loop itself, like Claude Code.

Tool calling

The model asks for a structured function call; the harness runs it and returns the result.

System prompt

Standing instructions that set the role, rules and boundaries before the conversation starts.

Prompting

Clear, specific instructions with examples and constraints, so the output is predictable.

Context window

The tokens a model can see at once; anything outside it does not exist for the model.

Context engineering

Choosing what goes into the window (files, docs, memories, tool results) and what stays out.

Short-term memory

What lives in the current session’s context and is lost when the session ends or is compacted.

Long-term memory

Decisions and facts saved outside the session, like Engram, and recalled when relevant.

Skills

Packaged instructions and scripts an agent loads on demand for a specific kind of task.

MCP

Model Context Protocol: a standard way to connect agents to tools and data sources.

A2A

Agent-to-Agent protocol: agents from different systems discover each other and delegate tasks.

Orchestration

Coordinating steps or agents so each one gets the right task, context and handoff.

Subagents

Child agents with their own context window, spawned for a focused task; they report back a summary.

Multi-agent systems

Specialized agents, such as planner, builder and reviewer, working toward one goal.

State graphs

Workflows modeled as nodes and edges with explicit state, so every run is predictable.

Routing

Sending each request to the right model, agent or tool based on its type or difficulty.

Evals

Repeatable tests that score agent output, so prompt or model changes can be measured.

Guardrails

Checks on inputs and outputs that block unsafe, off-policy or malformed results.

Sandboxing

Running agent actions in an isolated environment with limited files, network and permissions.

Human-in-the-loop

A person approves risky steps and reviews the result before it ships.

Observability

Tracing every prompt, tool call, token and cost, so failures can be debugged.

SDD

Spec-Driven Development: agents follow a written spec through plan, tasks, code and verification.

ODD

Organic-Driven Development: ship small, observe real use, then grow and refactor where needed.

Git worktrees

Separate working directories on their own branches, so parallel agents never collide.

Selected work

AIG4U

Insurance platform where the whole team books appointments, manages schedules, takes notes and coordinates in one shared workspace.

React
TypeScript
MUI

Visit AIG4U

Ten5

Voice-first fleet management: drivers use short commands to get information and leave comments while managers monitor and respond in real time.

React
TypeScript
Vercel

Visit Ten5

Buk

HR and payroll platform used by companies across Latin America. I built and maintained product features across the Rails backend and the jQuery front end.

Ruby on Rails
jQuery
JavaScript
HTML
CSS
Bootstrap
PostgreSQL

Visit Buk

Other projects

Full archive on GitHub

Circle Detection (opens in new tab)
Spots circular objects with OpenCV and signals an Arduino over serial to move a servomotor.
- Python
- OpenCV
- Arduino
Two-Wheel Robot in ROS (opens in new tab)
URDF/Xacro model with a caster wheel for balance, visualized in RVIZ.
- CMake
- ROS
- RVIZ
Web Interface for ROS (opens in new tab)
Controls the two-wheeled robot in real time from the browser.
- HTML
- CSS
- JavaScript
- ROS
Email Template (opens in new tab)
Responsive layout that renders consistently across major email clients.
- HTML
- CSS

Let’s talk

Have a project or a role in mind? Send me an email.

dario.cabascango.dev@gmail.com

DaríoCabascango

About me

Tech stack

UX engineering

Design systems

Accessibility

Interaction & motion

Verified in the browser

AI-native engineering

Agent orchestration

Skills & MCP

Quality gates

Concepts I work with

Agent core

Context & memory

Capabilities

Multi-agent

Reliability & safety

Workflow

Selected work

AIG4U

Ten5

Buk

Other projects

Circle Detection (opens in new tab)

Two-Wheel Robot in ROS (opens in new tab)

Web Interface for ROS (opens in new tab)

Email Template (opens in new tab)

Let’s talk