Skip to content / Saltar al contenido
SERVICE/02

AI Systems

Jorge Cortés puts AI where it changes the outcome, not where it looks impressive. Think names the decision. Build engineers the model into the product. Ship means production, not a notebook. WspWrapped is the case. The lab stays off-stage until it ships.

Jorge Cortés does the work, not an interchangeable team.

For whom

  • Founders who have a decision the product should make, not a slide that says “we use AI”.
  • Teams that need a pipeline — parse, compute, never store what it should not store — not a chatbot bolted on.
  • People who want the model behind a product someone can actually open.

What you get

  • The decision the system is allowed to make, named before a vendor is chosen.
  • A pipeline in the product: input, computation, the thing the user sees.
  • Privacy and storage rules that are architecture, not a policy PDF.

How I work

  1. 01/ Think

    Name the decision. If the product works without the model, the model is decoration. If it cannot, that is the work.

  2. 02/ Build

    Put the model where the outcome changes — an API when that is enough, a pipeline I own when the product has to remember nothing.

  3. 03/ Ship

    Production, languages, a free cut if the product needs one. WspWrapped ships reports and cards, not a research notebook.

Deliverables

  1. 01/A written decision: what the system computes, what it must never keep.
  2. 02/The live capability inside the product, not a separate “AI demo”.
  3. 03/The failure modes: timeouts, empty input, the user who pastes the wrong file.

Evidence

The public proof: a WhatsApp export in, 30+ metrics out, no message text kept: WspWrapped.

JORGECO/LAB holds experiments — Context Engine, Agent Sketchbook — that are not cases yet. They stay off this page until they ship.

Not for

  • “Add AI” with no decision that changes.
  • A wrapper around a chatbot as the whole product.
  • Automation for its own sake — ten steps and no one who owns the outcome.

Questions

Do you build with AI or just integrate APIs?

I use the model the product needs. Sometimes that is an API. Sometimes it is a pipeline I own — WspWrapped parses the export in memory and computes 30+ signals without storing message text. The test is the outcome, not the vendor.

How much does an AI system cost?

There is no published rate card. Price follows the decision the system must make, the data it is allowed to touch, and a fixed set of deliverables after a short diagnosis. Write through the contact page with that decision as you see it.

Do you work with remote teams?

Yes. Based in Santiago, Chile. Remote, worldwide. Spanish and English.

Do you do automation?

Automation is a means. If a flow removes real work — reminders, a parse step, an agent that verifies — it belongs in the system. If it is a stack of zaps and no product, this is not the work.

Do you store user data in a third-party model?

Not by default, and not as a way to skip architecture. WspWrapped never stores message text; only aggregates. What a given product may send to a model is a Think decision, written down, before Build.

What happens after launch?

The first production weeks are part of the work: empty files, languages, the query nobody predicted. That loop is scoped. It is not an unbounded retainer unless we name it.

Do I need my own model?

Almost never on day one. Own the pipeline and the decision. Rent the model until the product proves it needs something the API cannot do.

Name the decision