Gael Espinosa

Boring technology, on purpose

2026-07-10T00:00:00-04:00

Everything I ship lately is Django, server-rendered, HTMX where it needs to feel alive. No SPA, no build step, no client state to sync.

Not a limitation. A choice.

The receipts

CVora went from first commit to production in three days. Part of that is the stack having no moving parts I don't fully understand: Django views, HTMX partial swaps, one database, fly deploy. Every AI interaction on that site is a server round-trip that swaps a chunk of HTML. Nobody notices. Nobody cares. It ships.

This site: markdown files and Pelican. Builds in half a second. Hosting is free. A portfolio is text that changes a few times a year — a React framework here would be decoration.

The rule

New tech has to name the specific problem it solves for me, today, in this product. "We might need it later" doesn't count.

The part people miss: when later actually arrives, migrating away from the simple thing is easy — precisely because it's simple. Migrating away from the complicated thing you adopted early is the expensive move.

Not a religion

Sometimes the exciting tool wins. It just has to earn the spot with a problem I can point at, not a feeling that I'm falling behind.

The stack is not the product. Shipping is the product.

this blog — the site you're looking at

2026-07-10T00:00:00-04:00

What it is

The site you're reading right now. Everything on it — the experience timeline, the project cards, this very page — is generated from plain markdown files and one Python config.

The stack

Pelican — Python static site generator. Blog posts and project writeups are markdown files; the folder they live in (content/blog/ or content/projects/) decides which section they land in.
Tailwind CSS v4 + daisyUI 5 — all styling config lives in a single CSS file, no tailwind.config.js. The theme is one custom daisyUI theme: charcoal base, cream ink, hard green and loud yellow, zero border radius.
GitHub Actions → GitHub Pages — every push to main builds the CSS and the site, then deploys. No servers, nothing to maintain.
Formspree — the contact form on a static site, without publishing an email address anywhere.
ghchart — the contribution graph, flipped to dark mode with a CSS invert + hue-rotate trick.

Design notes

The look is deliberate neubrutalism: 2px borders, hard offset shadows with no blur, cards that physically press down on hover, Archivo Black shouting the headlines while IBM Plex Mono handles the labels. One theme, executed hard — no light/dark toggle to hedge with.

Why a static site in 2026

Because a portfolio is mostly text that changes a few times a year. Markdown in git, a build that takes half a second, free hosting, and nothing to patch on a Tuesday night. Boring technology, chosen on purpose.

Súbete Perra — local event discovery for Chile

2026-06-01T00:00:00-04:00

What it is

subeteperra.cl helps people discover events happening around them — music, arts, sports, tech, LGBTQIA+, food, and more — and lets organizers publish events without fees or gatekeeping. Browse by category, filter free events, reserve a spot, show up.

Fully bilingual (Spanish/English) via Django's i18n.

The stack

Django 6.0 on Python 3.13, dependencies managed with uv, everything running in Docker
PostgreSQL 17 behind the scenes, Tailwind on the front
Fly.io for hosting, WhiteNoise for static files, Sentry for errors
geopy powers the location-based discovery, icalendar exports events straight to your calendar
Claude API in the mix for AI-assisted features
django-axes rate-limits login attempts; MessageBird handles SMS

Engineering notes

Tested end-to-end — pytest + factory-boy for the unit/integration layer, Playwright driving a real browser for the flows that matter (browse → register → attend).
A companion Chrome extension lives in the same repo for capturing events from around the web.
Boring, deployable architecture — one Django app, one database, one fly deploy. No microservices for a product that fits in one head.

ClickDocs — export ClickUp docs as markdown

2026-05-01T00:00:00-04:00

The problem

Teams write years of documentation into ClickUp, and then it lives there — locked behind the app, unsearchable from the terminal, invisible to git, and awkward to migrate. ClickDocs pulls it out: point it at a workspace and get clean markdown files back.

How it's structured

Two components in a UV workspace:

clickup-exporter — a standalone Python library and CLI that walks the ClickUp API (docs, nested pages) and writes them out as markdown. Small dependency footprint: requests for the API, mdutils for generation.
export-site — a Django + Tailwind web UI wrapping the library at clickdocs.xyz: register, connect your workspace, export. Freemium — the first 10 pages are free.

Deployed on Fly.io with Docker.

Interesting bits

Nested page trees — ClickUp docs are hierarchical; mapping that onto a sane directory structure with correct relative links is most of the actual work.
API testing without the API — the test suite uses pytest-vcr to record real ClickUp API responses once and replay them forever, so tests are fast and deterministic.
Library-first design — the web product is a thin wrapper; anyone who'd rather stay in the terminal can use the CLI directly against their own API token.

CVora — AI-powered career toolkit

2026-03-01T00:00:00-03:00

What it does

CVora takes your resume and puts it to work against real job descriptions:

ATS scoring — paste a job description, get a match score with a breakdown of what lands and what's missing.
Batch comparison — score one resume against multiple job descriptions at once and see where you're strongest.
Per-role coaching — an interactive session that walks through your resume in the context of a specific role and suggests concrete improvements.
Structured export — resumes live as RenderCV YAML, so output is versionable and renders to a clean PDF for download.

How it's built

Django + HTMX on the backend-driven-UI end, the Claude API for the scoring and coaching intelligence, deployed on Fly.io. From first commit to production in three days.

The HTMX approach keeps the whole thing server-rendered: no SPA build step, no client state to sync — every AI interaction is a partial swap.

ink+ — create Inkdrop notes from Alfred

2024-09-26T00:00:00-03:00

The itch

The official Inkdrop Alfred workflow could only search notes. Capturing a thought still meant switching to the app, finding the right notebook, making a note — enough friction to lose the thought.

The contribution

Merged PR #4 adds an ink+ command: type ink+ title: contents and the note lands in Inkdrop without ever leaving the keyboard.

What went in:

ink+ <title>:<contents> — new command hitting Inkdrop's local server API to create the note.
defaultNotebook and defaultTags workflow environment variables, so captured notes are filed and tagged automatically.
Refactor — search.php became inkdrop.php, since the workflow was no longer just search.
Failure feedback — a macOS notification when the local server errors out, instead of silently eating the note.

Why it's on this list

Small diff, daily payoff — this is the kind of tooling contribution I enjoy: find the friction in a tool you actually use, remove it upstream so everyone gets the fix.

Inkdrop linking plugins — note-from-selection + search-notes

2023-05-13T00:00:00-04:00

One workflow, two plugins

Inkdrop is a markdown note app; out of the box, linking notes together meant hunting for IDs by hand. These two plugins cover both directions of a wiki-style linking workflow:

note-from-selection — selected text becomes a new linked note
search-notes — link to (or jump to) an existing note, searched from inside the editor

note-from-selection

Select text, hit cmd-shift-e, and the selection becomes its own note — the original text replaced by a link. cmd-shift-ctrl-e pops a notebook picker to file it elsewhere. The details that matter:

No naming prompt — the first line of the selection (up to a configurable 80 chars) becomes the title, so splitting a note never breaks flow.
Links that stay fresh — the inserted link is left with blank text on purpose: Inkdrop then renders the referenced note's live title, status, and tags in previews. The title also lands in an HTML comment next to the link for readability in the raw editor:
```
[](inkdrop://note:<noteId>)  
```

search-notes

The second plugin opens a search modal without leaving the editor:

cmd-shift-s — search notes and insert a link to the selected result; works on selected text too.
alt-cmd-space — search and navigate to the result, which pairs naturally with Inkdrop's back/forward navigation commands.

Coexisting with the core app

Inkdrop v5.8.0 later shipped its own note-from-selection and backlinks panel. Instead of dying, the plugins leaned into what core doesn't do — prompt-free default titles, cross-notebook filing, link-insertion from search — and dropped the features core now did better. A plugin shouldn't compete with the app on things the app does well.

Virtualenv env vars

2022-07-08T00:00:00-04:00

Setup postactivate script on a new virtualenv

`~/.virtualenvs/project/bin/postactivate`

#!/usr/local/Cellar/bash/5.0.18/bin/bash
PROJECT_PATH=~/some/project/path/

export $(grep -v '^#' $PROJECT_PATH/.env | eval echo $(eval xargs))

`~/some/project/path/.env`

# DEV ENV
# DATABASE_URL=<some_dev_env_db_url>

# LOCAL ENV
DATABASE_URL=postgresql://<local_db_url>

grep of all not commented lines allows for an easy env vars switch.

With virtualenvwrapper, just the .env file and rerun the workon project cmd.