AI & Automation Lead at EPA.Digital · Mexico City · since 2017
I build AI for marketing people. And for finding an apartment.
I’m AI & Automation Lead at EPA.Digital, a marketing agency in Mexico City. 28 people on the account teams use the media MCP server I started by myself in 2025. On my own time I build agent apps: one to plan trips, one to find an apartment and one to find a job.
A few numbers
- active people on the media MCP server, with over 14.7M tokens
- 28
- Source: admin dashboard, Oct 2026
- for Coppel’s daily SKU analysis, down from a full day
- 30 min
- Source: Coppel project, 2026
- people, the whole agency at the time, with access to the Prompt Library
- 170
- Source: launch, Nov 2025
- records loaded into BigQuery by serverless ETLs with retries and backfill, for Tickspot and Coppel
- 420,150
- Source: Tickspot and Coppel ETLs, 2025
Four projects to start with
- 484real conversations reviewed to measure quality
An assistant that queries Meta, Google Ads, GA4 and three more channels for account teams
EPAIn productionRead the case - v1what I learned when the first version went unused
The agency’s knowledge base. Not even we used the first version
EPAIn productionRead the case - 0unverified places in an itinerary
Plans the trip with you and checks every place before proposing it
PersonalDeployed · personal useRead the case - CDMXevery area measured with open FGJ and DENUE data
Hunts for apartments in Mexico City, measures the area and distrusts what looks off
PersonalIn progressRead the case
Other things I do
- 12 projectsWork at EPAAI tools for the agency’s teams, plus the rules and courses to use them.Media MCP server · Dashboards and Hopper
- 5 projectsOwn productsApps I build after work, with AI at the core and the app built around it.Iterum · Domi
- 4 projectsData and analyticsMy work as Data Manager from 2020 to 2024: BigQuery, R and a lot of reports.Central media database · Client analyses in R and Quarto
- 2 projectsTeaching and talksInternal courses and a GEO course at Universidad Panamericana.Internal courses · GEO course at Universidad Panamericana
- 3 projectsLabSmall tries to learn something. I’ve dropped a few.Office game · Spend analyzer with local Llama
Four rules I reuse in every project
- 01
Plan before coding
It’s the most boring part and the most important: sitting down before the code to agree with others on what’s worth building. It’s what keeps you from projects nobody uses.
I learned it with Einstein - 02
Propose, don’t impose
I learned it leading CRO: for A/B tests I opened the discussion to the team instead of imposing my idea, and often theirs was better. In Iterum, if an agent wants to change something, it proposes it with Accept and Reject.
Example: Iterum - 03
Every figure has a source
In Domi, every price says which site it came from and when it was checked, and every crime figure gives its period and FGJ dataset.
Example: Domi - 04
Not everything is an LLM
In Hopper, checks against the data are plain code. Gemini only reads the context the analyst writes.
Example: Dashboards and Hopper
From accounts to AI, year by year
- 2017Account executive; my first automated reports
- 2018Digital marketing analyst; learned R from a newsletter
- 2019CRO & Data Analytics Lead, team of 3
- 2020Data Manager
- 2024AI & Automation Lead; Verne, my first AI tool
- 2025First MCP servers
- TodayMedia MCP server, Kalman, Hopper and my own products
I’m José Carlos Corona.
If you’re hiring for an AI Engineer or Applied AI role in Mexico City, message me on LinkedIn.
Open to AI Engineer and Applied AI roles · Mexico City, hybrid or remote