Who you would be working with
Datasmarts is one person: Jesus Martinez, a software and AI engineer with 14+ years of experience and 10+ years shipping production machine learning systems. You talk to the person who does the work, which is the main thing a small consultancy has to offer.
I did not arrive at AI from the marketing side. I spent the first part of my career on back-end systems, the unglamorous work of moving monoliths onto services, building ingestion that does not lose records, and making queries fast enough that people stop avoiding the report. That background is the reason I ask what your data looks like before I talk about models.
The machine learning work came next, and for most of a decade it was production systems rather than notebooks: classification and extraction pipelines, forecasting, computer vision, the parts that have to keep running after the demo. Language models changed what those systems can do, not the discipline around them. An accuracy bar agreed in advance, a way to trace an answer back to its source, and a person in the loop where being wrong is expensive are all older ideas than the current wave.
Datasmarts exists because the interesting problem moved. Most businesses do not need a model built. They need someone to work out which of their processes is quietly costing them a week a month, and then build the thing that removes it. That is a consulting problem with an engineering answer, and it is the work I want to be doing.
How I got here
- The first years
Back-end and distributed systems
Monolith to microservices migrations, event-driven architecture, and the first CI/CD pipelines at more than one company. Python became the primary language here and has stayed that way for 12+ years.
- 10+ years
Production machine learning
Classification, extraction, forecasting and computer vision systems built to run in production rather than to demo. This is where the habit of agreeing an accuracy bar before the build comes from, because that is the only version of the work that survives contact with real data.
- Recent years
LLM systems and workflow automation
Retrieval architectures, document extraction pipelines, multi-agent systems, and self-hosted orchestration with n8n and Python. The engagements behind the case studies on this site sit here.
- Now
Datasmarts
Consulting on which process to automate, then building and measuring it. Working with clients in the US and EU remotely, in English and Spanish.
Whether this is a fit
Worth being direct about, because a wrong fit wastes your budget before it wastes my time.
This works well if
- You have a specific process that repeats, consumes real hours, and someone can walk me through it
- You would rather know a process is not worth automating than pay to find out afterwards
- You want the result measured against the number it started at, not described as an improvement
- You need the work to keep running after handover, on your infrastructure and your accounts
- You want to be talked out of the expensive option when the cheap one is correct
This is not a fit if
- You are looking for staff augmentation. Datasmarts is not a body to take tickets or extend an in-house team on someone else's plan.
- You have budget allocated to "do something with AI" and no particular process in mind. The assessment would most likely tell you to spend it elsewhere, and you can reach that conclusion without paying for it.
Tell me which process is costing you the most
That is the whole first conversation. If it turns out not to be worth automating, I will say so.