I’m Brian. I run American Data Networks in Costa Rica. Here I share what we’re building, the problems we run into, and what I’m learning along the way.
TelecommunicationsPractical AIBusiness automationLearning by doing
The story behind the work
We decided to build it ourselves.
I spent two years waiting for the automation I wanted. Then I told my son we needed to do this ourselves. We had a lot to learn.
Our AI journey · A personal account
We had to learn. So we got started.
Mike’s question about installing Linux is a moment I want to preserve. So is José Alejandro’s willingness to take on anything—from hardware to software. This is their story too.
The starting point, as I remember it. I’m still putting the exact timeline together.
What needs to work better? 01
What do we already know? 02
Can we build something useful? 03
Did it actually work? 04
These questions keep us moving. We are still working through the answers.
Selected projects
The problems we’re working on.
Each of these started with something I wanted to work better in the company. Some pieces are working. Others are still being tested or put together.
01 / COMPANY KNOWLEDGE
American Data Brain
Built a foundation · Still improving it
We had information in emails, customer records, and work orders. I wanted us to be able to find it. Brain brings those sources together, but gaps and conflicting records still need attention.
I want a customer’s question to reach the right place. We built a router to help with that. Fresh tests showed us wording it still misunderstands, especially when people ask in unexpected ways.
Finding information is one thing. Turning it into a useful reply is another. We’re working on how to draft an answer, check it, and get it back to the right conversation.
I want to know who has taken responsibility and what happens next. We built a dashboard to bring conversations and commitments together. A closed ticket still isn’t proof of a solved problem.
Checking records · Keeping unclear cases in review
The service record and the bill should agree. We built an auditor to spot differences and make only narrowly approved corrections. When the match is unclear, it stays with a person.
We tried a local AI model on real email passages. It answered quickly, but some answers added facts that weren’t there. Finding that before staff relied on it gave us a clear next problem to solve.
I’m the CEO of American Data Networks. A lot of my questions start with everyday work: why does this take so long? Why is someone doing it by hand? What would it take to make it work better?
This site is where I want to share what happens when we try to answer those questions. The AI journey is a big part of that. I want to show the people, the progress, and the things we haven’t figured out yet.
