01§
Workflows between your tools
Forms, CRM, spreadsheets, email, WordPress and your database talking to each other on a schedule or on a webhook, built in n8n, or in TypeScript when the logic outgrows it.
AUTOMATION · n8n & AI AGENTS
That process eating hours of your week, running on its own.
I connect the tools you already use, add AI where it saves real work, and leave the result running in your own accounts. Before I build anything you approve a diagram of the flow: what it touches, what it decides and what happens when something fails. After launch, it alerts you when it breaks or stops running, instead of failing in silence.
What I automate.
02§
AI agents that do the work
Classify, score, summarize and rewrite with a language model, with a cheap model filtering first so the expensive one only sees what is worth it.
03§
WhatsApp and voice
Assistants that answer your customers from your own information and book against your real calendar, without letting the model invent a time slot.
04§
Nothing fails in silence
Alerts when a workflow fails or stops running, drafts instead of auto-publishing wherever a person should approve, and a heartbeat outside your server.
Four steps, the diagram comes first.
- 01
Map
A 30-minute call about the one process that costs you the most time: where the data comes from, who touches it and where it has to end up.
- 02
Diagram and scope
The flow drawn step by step, with the systems it touches, the decisions it takes and what happens when something fails. You approve it, with the scope, before the build starts.
- 03
Build
Same weekly rhythm as every project: one meeting with a demo and one delivery every Friday, tested with your real data, not a sample.
- 04
Launch and handover
Alerts set up, the workflow running in your accounts and enough documentation for anyone on your team, or another developer, to maintain it.
Automations I have built.
News pipeline in n8n
Built for a friend’s news siteReads around twenty RSS feeds every three hours, drops repeats in Postgres, scores each story with a cheap model, rewrites only the relevant ones with a stronger model and leaves them as WordPress drafts with a cover image and a Telegram alert. A person approves before anything is published.
TechBlog AI Agent
Own project · archivedDual-agent system that discovered AI and tech news from 20+ RSS feeds, rewrote it in Spanish and published it to an Astro blog every three hours from a cron inside Docker, with PostgreSQL-backed deduplication and its own Git commits.
Voice agent that books appointments
Proof of concept · public repoA visitor talks to a page in Spanish and hangs up with a confirmed 30-minute appointment on the business calendar. The backend checks real availability in Cal.com, so the model cannot invent a slot or book the same one twice.
Clona
Client projectB2B platform for AI conversational agents that answer from each client’s documents through vector search, delivered across chat, voice, WhatsApp and embeddable web widgets.
When I would not automate it.
It happens a couple of times a week
Building and maintaining the workflow costs more than doing it by hand. I will tell you so on the call.
Nobody can review the output
If the AI writes something that goes out under your name and no person approves it first, sooner or later something wrong gets published.
The process is not defined yet
If every case is decided differently, automating it only makes the confusion faster. We define the rules first, then automate them.
FAQ
Questions people actually ask.
Which process should I automate first?
The one that repeats every day or every week, follows clear rules and eats the most hours of someone on your team: copying data between tools, answering the same customer questions, preparing a report, moving leads into a CRM. Ramón starts every automation project with a 30-minute call about that single process and a diagram of the flow, so the first workflow pays for itself before the next one is planned.
Do you build automations with n8n?
Yes. Ramón uses n8n for most workflows that connect existing tools: RSS, databases, WordPress, email, Telegram and HTTP APIs, on a schedule or on a webhook. His news pipeline in n8n reads around twenty feeds every three hours and leaves WordPress drafts. When the logic outgrows a visual workflow, as with TechBlog AI Agent, he writes it in TypeScript instead.
What happens when an automation fails or stops running?
It tells you. Every workflow Ramón delivers has alerts for failed runs, and for the case n8n does not report on its own: a workflow that is switched off or a server that is down produces no failed run at all. For that he uses checks on when each workflow last succeeded and a heartbeat to an external monitoring service, so the alert arrives even if the whole server stops.
Can you automate WhatsApp or phone conversations with AI?
Yes. Ramón worked on Clona, a client platform whose AI agents answer from each customer’s documents across chat, voice and WhatsApp, and built a public proof of concept of a voice agent that books 30-minute appointments in Cal.com. In both, the model answers from real data and the backend makes the decisions that must not be invented, such as available time slots.
Do I get to approve the automation before it is built?
Yes. The second step of every automation project is a diagram of the flow: the systems it touches, the decisions it takes and what happens when something fails. You approve it together with the scope before any building starts, and from then on you see progress in a weekly meeting with a demo and a delivery every Friday.
Who owns the workflows and accounts once the project ends?
You do. The workflows run in your accounts, the code lives in your repository and the documentation is enough for someone on your team or another developer to maintain it. Nothing depends on Ramón staying on the project.
SECOND OPINION
When you need judgment,
not just code.
Not every engagement needs someone to take the whole product. Sometimes you already have a team and need someone who has shipped this before to look at it. A decision nobody can settle, a stack under evaluation, a roadmap that feels wrong, an estimate you do not trust. A decade of shipping across web, mobile and AI left me something more useful than a stack: judgment about what is worth building.
