FernandoApplied AI

Fernando

Applied AI lead

I build AI agents and the apps around them, from the first conversation about where AI fits to the version people use every day.

Before AI I managed a small business, so I know first-hand the problems companies face.

See my work
Agents

Most of my work is building agents.

Assistants that take actions. Some work on their own and some check with a person first, depending on what the task needs.

  1. The right amount of autonomy

    Not every agent needs someone to approve its work. I set how much each one does on its own by what a mistake would cost, and widen it as the agent proves itself.

  2. One agent for email, chat and phone

    The same agent answers across channels and hands the conversation to a person when it should.

  3. Answers from a reviewed knowledge base

    The agent only answers from content the team has checked. When it can't answer something, that question goes on the list of things to add.

  4. An assistant for each person on a team

    It reads, drafts and takes routine work off their plate. What it may do without asking is something each person sets.

Testing and evals

I don't ship anything I haven't tested on real data.

How I know whether a model or a prompt is good enough, and what it costs.

  1. Testing on real data

    I run the whole set of real inputs instead of a few demo cases. That's usually where the problems nobody reported show up.

  2. Automated grading I can trust

    I check automated scores against reviews by people. When they disagree, it's often the rubric that needs fixing, not the model.

  3. Cost tracked with quality

    Every test run records what it cost, so switching models or prompts is a decision with real numbers behind it.

  4. Reading transcripts

    I read actual conversations to find the questions an agent keeps getting wrong, and I fix those first.

Apps

Sometimes the AI needs a whole app around it.

Web and mobile apps, from the first prototype to launch.

  1. Web and mobile apps

    I've built and launched apps for web, phone and tablet, and kept them running after launch.

  2. Payments

    Subscriptions, tiered pricing and in-person payments, built in from the start.

  3. Apps for many customers

    Each customer gets their own space, with roles and properly separated data. I do a security review before the first one signs up.

  4. Design systems

    Each product gets one set of colors, type and components, so the web app, the mobile app and the documents all look like the same product.

Data and analytics

A report is only useful if it's current.

Getting current numbers in front of the people who make decisions.

  1. Live dashboards

    Dashboards that pull straight from the database, so you're looking at today's numbers and not last week's export.

  2. Tracking checks

    Scheduled checks catch broken tracking, fix what they're allowed to fix and flag the rest.

  3. Data questions in plain language

    I connect data sources to AI assistants so people can ask questions that used to need an analyst.

  4. Automating systems with no API

    When a system has no API, a browser agent keeps it in sync and logs every change it makes.

Tools for teams

Some of the most useful work is making a team faster.

Packaging what a few people know so the whole team can use it.

  1. Plugins and skills

    I turn what a couple of experts know into tools the rest of the team can run: checklists, reference data, troubleshooting steps.

  2. Overnight agents

    Agents run on a schedule overnight and leave reports and drafts for the team in the morning.

  3. Documentation

    Help docs that live next to the code, with a screenshot for each step, so they get updated when the product changes.

  4. Teaching

    I teach developers how to get AI code ready for production.

About

Technology should make things easier.

I'm Fernando. I'm passionate about technology, and I believe it's the way to improve people's lives.

Before I worked in AI, I managed a small business. It showed me the struggles and challenges that companies face, and it lit something in me: don't stop there, keep looking for solutions.

I'm still amazed by how much easier technology has made things for me, and I'd love to share that experience with as many people as I can.

So, let's chat!

Writing

The longer versions

One project from start to finish, and how this site is put together.

Case study

A front-desk agent for a small hotel

A small hotel was missing half of its phone calls. What I built, how I check it, and what a review found.

Read Design

The design system behind this site

Colors, type and components I built for my own projects, with the moving background to play with.

Read
Lab

How this site is built

This site runs on a design system I made for my own projects: one set of colors, type and components. It borrows from 70s and 80s audio gear, with dark, warm panels, amber and teal lights, and monospace labels.

See the design system

Background

The moving lines

The lines behind the page are drawn on a single canvas. They drift like tape in the wind, go calm under text and ripple when you change sections. Move your mouse over them.

The wind is real. It replays, hour by hour, the wind over Mexico City, where I was born, on the day I was born. Each hour of that day lasts twelve seconds here. The wind's speed sets how hard the lines are pushed, and its direction sets which way the gusts cross, with north at the top. Lift is how high the tallest line rises, in pixels.

Weather data: Open-Meteo

Contact

Want to work together?

Send me a note about what you're working on. I read and answer everything myself.

Privacy

01/ 10
Scroll, swipe or use the arrow keys