Independent builder · Senior frontend engineer

Hey, I'm David Minguela.

I turn difficult workflows into useful software.

Frontend architecture, product systems and the infrastructure needed to ship them.

I work where product ambiguity meets technical complexity — designing the interface, the system behind it and the path to production.

Working across

  • Product · Systems
  • Frontend · Architecture
  • AI · Workflows
  • Owned · Infrastructure

Selected work / 01—06

Software with a job to do.

A portfolio of real systems: private tools, public products and infrastructure built around decisions, not demos.

Case 02

Commerce workflow

Preview available

Cesta Carrefour

From an ordinary shopping list to a cart a person can actually review.

A Nuxt application and Chromium extension that resolve product candidates, remember decisions and hand a reviewed list to an existing Carrefour session.

The friction
Product matching is ambiguous, retailer pages are volatile and checkout must remain explicitly human.
The system response
A layered review workflow with safe fallbacks, remembered choices and a deliberate extension handoff before purchase.
Open preview
  • Nuxt
  • Vue
  • PGlite / Neon
  • Chromium MV3

What makes it real

  • Nuxt Layers
  • MV3 extension bridge
  • Human-in-the-loop

Case 03

Nutrition workflow

Public product

Menu Planner

Nutrition planning that makes documents and AI useful without making them the interface.

A planning product that connects OCR, AI workflows, Supabase and Vercel while keeping fallback paths visible and usable.

The friction
Unstructured documents and unreliable upstream services create friction at the exact moment a user needs a clear plan.
The system response
A product flow that turns document input into editable planning data and degrades safely when automation is unavailable.
Open product
  • Nuxt
  • Supabase
  • OpenAI
  • Docker

What makes it real

  • OCR ingestion
  • AI-assisted planning
  • Resilient delivery

Case 04

Decision support

Local-first tool

Car Finder

An evidence workspace for comparing cars without losing the source behind a claim.

A local-first tool for importing listings, curating technical evidence and publishing structured comparisons for people and AI assistants.

The friction
Listings disappear, equipment claims conflict and important buying decisions end up based on fragile browser tabs.
The system response
A source-aware catalogue with duplicate review, cautious evidence states and public Markdown and JSON views.
Local tool
  • Node.js
  • SQLite / Neon
  • HTML extraction
  • JSON-LD

What makes it real

  • Source preservation
  • Duplicate review
  • AI-readable exports

Case 05

Private infrastructure

Private system

NAS Gateway

A private file workspace where the filesystem stays the source of truth.

A Nuxt interface over NAS storage with stable identities, search metadata, favourites, trash and activity history.

The friction
A useful file interface needs product-level metadata without pretending a database owns the underlying files.
The system response
A split-responsibility architecture: bytes remain on disk while SQLite carries searchable, recoverable product state.
Private system
  • Nuxt
  • SQLite
  • Authentik
  • Docker

What makes it real

  • Filesystem authority
  • Stable metadata
  • OIDC access

Case 06

Personal operations

Product in evolution

Renovaciones

Renewals and subscriptions organised as a timeline, not another forgotten spreadsheet.

A cross-platform product for tracking renewals, recurring costs and notification channels across web and mobile.

The friction
Insurance, subscriptions and licences live on different cycles and usually become visible only when they charge.
The system response
A single operational timeline with cost views, cloud sync and configurable notification paths.
Open product
  • Expo
  • React Native
  • Supabase
  • Vercel

What makes it real

  • Web + mobile
  • Cost overview
  • Notification workflows

Operating principles / 04

How the work holds together.

The interface is only one layer. I shape the product boundary, the data and the operational path around it.

01

Reduce ambiguity first

Before choosing components, make the decisions, actors and failure states of the workflow visible.

02

Design the boundary

Keep product rules, infrastructure and interface concerns separate enough to evolve without theatre.

03

Ship the whole path

A good screen is not delivery. Auth, data, fallbacks, mobile use and deployment belong to the same problem.

04

Own the operation

Observe the system, document the decisions and make recovery a product capability rather than an afterthought.

Working toolkit

Tools organised by responsibility.

Technology choices follow the system. These are the pieces I currently reach for to design, ship and operate it.

01

Shape

Product language and interface systems.

  • Vue
  • Nuxt
  • React
  • Next.js
  • TypeScript
  • Tailwind

02

Persist

Data that survives the happy path.

  • Postgres
  • Neon
  • Supabase
  • SQLite
  • Drizzle

03

Connect

Automation, maps and intelligent inputs.

  • OpenAI
  • OCR
  • n8n
  • Mapbox
  • Browser extensions

04

Operate

The path from repository to a running system.

  • Vercel
  • Docker
  • Traefik
  • Cloudflare
  • GitHub Actions

Build log

No fake telemetry. Just the work.

Short notes on what I am building, changing and learning — including the awkward parts that polished case studies usually remove.

NOW.01

AI as a workflow participant

Interfaces where automation proposes, people decide and the system preserves why.

NOW.02

Evidence-aware products

Turning documents, pages and changing sources into claims users can inspect and correct.

NOW.03

Small systems with real operations

Taking focused tools all the way through access, persistence, observability and recovery.

Open channel

Have a difficult workflow?

I can join a product team or help turn an operational problem into software people can rely on.

For product teamsFor focused builds