Permissions per tool, not a master key
The AI reaches only the folders, sheets and calendars you authorize, one by one, through dedicated service accounts. What isn't authorized doesn't exist for the system.
AI consulting · Querétaro, Mexico
People decide. AI executes.Your company, smarter every month.
Putting AI into a company is not bolting on a chatbot. It is mapping how the company actually works — processes, people, brand, data — and connecting everything connectable: WhatsApp, email, Drive, Shopify, your ERP, custom APIs.
- Claude Partner Badge · Anthropic
- Whole-company mapping
- Custom flows and integrations
- Permissions and audit log
Hover or tap a system to see what it connects.
The point
It matters. It just doesn't need a person.
Logging an order, putting a quote together, moving data from an inbox into a spreadsheet, sending the appointment reminder. All of it matters — skip it and the operation falls apart. But none of it needs a person's judgment. And while your team is doing that, it isn't doing the work that does.
Deliverables, not data entry
Your team's hours go into producing the work clients pay you for, not into moving information from one place to another.
People with people
The hard conversation, the negotiation, the angry client, the decision with incomplete information. No automation replaces someone from your team there.
The tedious part, done just as well
An automation doesn't get distracted and doesn't forget at six on a Friday. The routine work doesn't just leave your desk — it comes out more consistent.
The ecosystem
Own your processes, not hostage to a platform
What makes an AI useful is not the model: it is the context and the processes you give it. That part is yours, and it moves with you.
- AI — the processor
- Reads your documents, understands your company's context, drafts, classifies and decides the next step. Today Claude plays that role.
- Flow — the connector
- Moves information between systems, triggers the processes and enforces the permissions. Today it runs on n8n. It is the hard part, and it is the part I build.
Today the processor is Claude and the flow runs on n8n. Tomorrow they could be something else: if your processes are defined and documented, they move with you without redoing the work. What you are buying is not a platform subscription.
Documents that generate themselves
Quotes, receipts, contracts, reports and branded stationery, assembled from the data already sitting in your sheets and your Drive.
Processes that talk to each other
A Shopify sale moves inventory, triggers the invoice, notifies on WhatsApp and shows up in Monday's report. With nobody copying and pasting.
A system that evolves
Every connected process leaves data that reveals the next bottleneck. The ecosystem isn't delivered finished — it's delivered growing.
Method
Understand the company first. Connect it second.
Nobody automates well a process they don't understand. That's why technology is the fourth phase, not the first.
- 01
Mapping the company
Internal and external processes, people and who actually does what, brand and stationery, documents, data, systems. The full briefing, not the org chart.
- 02
Deep research
Your competition, your market and your own data, analyzed in depth. This is where AI earns its keep: seeing what nobody inside can see — where money is leaking and what the company next door is doing.
- 03
Designing the ecosystem
What connects to what, in what order, and where each human control point sits. The output is a map of the full network with its permission architecture, before a single line of code.
- 04
Building the connections
This is where n8n, custom APIs and the integrations that don't come out of the box come in. It's the hard part of the work and the part I take on, with tests and an audit log from the first flow.
- 05
I teach you to use the AI
I sit down with your team and teach them to actually work with AI, not to copy prompts. The goal is that the thinking is yours and only the hard part depends on a specialist.
- 06
The ecosystem keeps growing
With the network running and measured, the next process to connect stops being a hunch. It's decided with the data the system itself is already producing.
One concrete example
From a message to a confirmed appointment, with nobody typing
One thread of the network, so you can feel how it works. A real operation has several running at once.
- 01
The client arrives
A prospect writes on WhatsApp or sends an email. No form to fill in, no new system your team has to learn.
- 02
It records itself
The details land in your master sheet: who they are, what they asked for, which channel they came through and when. The client lost between conversations stops existing.
- 03
The AI reads only what you authorized
Access to specific Drive folders, not to all of Drive: past quotes, price lists, client history. It answers with your company's information, and what it consulted is logged.
- 04
It books and documents
It finds the open slot, respects the team's schedule, creates the appointment and generates the quote using your template and your brand.
- 05
It confirms on the same channel
The client gets the confirmation on WhatsApp, where they were already writing. If they reschedule or cancel, the calendar and the sheet update without anyone touching them.
- 06
It reports back to you
Every morning you get the summary: today's appointments, what's pending, who hasn't replied. Every week, what's getting stuck and at which step.
Automation without governance
- All-or-nothing permissions over your Drive
- No record of what the AI did
- The vendor holds the keys
- If something goes wrong, nobody can reconstruct it
- Your client hits a bot that knows nothing about them
Governed automation
- Permissions per tool and per folder
- An audit log of every action, with time and actor
- The access belongs to your company
- Human approval where the mistake is expensive
- It answers with the client's real history
Built on top of what already runs
- WhatsApp Business
- Gmail
- Google Drive
- Sheets
- Calendar
- Docs
- Microsoft 365
- Shopify
- n8n
- Claude
- OpenAI
- Custom APIs
Data governance
What your legal team is going to ask
And the reason most of these projects stop right there.
An audit log of every action
Every read, every appointment created, every row written is recorded with time, actor and result. Auditable months later, not just on demo day.
Human approval where the mistake is expensive
The AI proposes, a person approves. Exactly where that control point sits is defined with you before anything gets built, and it's put in writing.
Personal data masked in the logs
The logs don't store personal data in the clear. It's the same practice I apply in systems that move money, where a leak carries legal consequences.
The access belongs to your company
Credentials, service accounts, repositories and documentation are in your name from day one. If you change vendors tomorrow, you lose neither the system nor control of it.
Regulatory context, without the sales pitch
Mexico's data protection law, in force since March 2025, does not explicitly address AI processing. That doesn't reduce your exposure: it means the design has to stand on its own. In the diagnosis we review what purposes your current privacy notice declares. I don't replace your legal counsel; I give them something concrete to review.
Services
Four ways in
Most engagements start with the first or the third.
Designing and connecting the ecosystem
The full job: mapping, network design, integrations with n8n and custom APIs, automated documents and templates, permissions and an audit log. It grows in phases, not all at once.
Deep research and analysis
Competition, market and your own data, analyzed in depth with AI. The output is a document with what you weren't seeing and a prioritized list of what to attack first.
An audit of the AI you already have
You already have a bot or automations running and nobody knows what they can reach. I review permissions, close what's open beyond need, install an audit log and hand you the map of what touches what. It's the cheapest work on the list and the one that removes the most risk.
Training so you don't depend on me
I sit down with your team and teach them to use AI in their actual work, with clear rules of use. What I do keep is the integration layer, because that's where everything breaks if it's done badly.
Pricing
What it costs, before we talk
Publishing the price should be normal. Almost nobody in Mexico does it.
First conversation
No cost30 minutes
You leave that call knowing whether AI applies to your operation. If it doesn't, I tell you then and there's no commercial follow-up.
Diagnosis of one area or process
from $38,000 MXNfrom 3 weeks
You stop guessing where to start. One area, what it costs to run today, and the first process to connect with its success criteria in writing. Credited in full if you continue to the project.
Whole-company mapping
from $85,000 MXNfrom 5 weeks
You see your company as a system, not as separate areas. The output is the map of everything connectable, ordered by what costs most today, and the phased plan to work through it.
One connected process
from $45,000 MXNfrom 4 weeks
One process stops needing someone typing it. It runs end to end, with permissions and an audit log, and your team is trained to operate it. This is where most engagements start.
Connected ecosystem
from $120,000 MXNfrom 8 weeks
Your processes stop sitting in isolation and start talking to each other. Documents that generate themselves, an operations dashboard, and reports that arrive without being asked for. Handed over documented to your team.
Ongoing support
from $18,000 MXN/monthoptional
The ecosystem keeps growing every month without you pushing it. It ships ready to run without me; this is for when you'd rather someone keep expanding it.
Every figure is a floor, not bait. Publishing an invented ceiling would be worse than publishing nothing, because real scope moves it: how many processes, how scattered your data is, how many systems have to be touched and how often they change. On the first call we bound it and I give you the number for your case in writing. And if it doesn't fit this quarter's budget, it gets split into phases: we start with the one that pays for itself and the rest waits.
- Staged payment
- 50/50 on short projects, 30/40/30 on long ones. You don't pay everything up front.
- Three months of technical warranty
- Adjustments included after delivery. If something breaks in that window, I fix it at no cost.
- Success criteria in writing
- Defined before any code is written. If the pilot misses them, it stops there and the next phase isn't billed.
- Everything stays in your name
- Access, credentials, documentation and training. If we finish, the system keeps working without me.
Who's behind it
Derian André Castillo Franco
Engineering lead, fullstack engineer and designer. Founder of XenthAI.
You don't learn this from watching videos. AI that actually works inside a company is built with the same discipline as any system that survives production, and that discipline only comes from having carried one.
Ten years building software for Mexican companies: construction, fuel distribution, real estate, education, sports and foundations — more than 35 projects delivered since 2016, with clients ranging from the family business to the international corporation.
Today I'm a Senior Fullstack Engineer at a Big Four firm, where I technically lead an international team. That's why I can tell you precisely what that tier won't do for a company your size: it doesn't publish a price, it won't give you a pilot with fixed scope and duration, it won't tell you who specifically will be in your office on Tuesday, and its case studies are global corporations, not Mexican companies of 800 people. That work is excellent for whoever needs it. You probably need something else.
I'm also Engineering Lead of Sul Finance, a financial platform in production, and of EnXingaPay, a crypto wallet product I took from architecture through to visual identity. That's where I learned to design systems that fail closed, that record every movement, and that someone else can operate.
And the proof that actually matters: the AI engineering infrastructure I built for my team was adopted by the entire team, and 85% of commits ended up AI-assisted, verifiably. Not a pilot that got presented in a meeting and died. This is AI in real teams, real infrastructure, and projects where a mistake costs money.
- Process first, AI second
- A good share of what looks like an AI problem is a badly defined process. Fixing it there costs less and returns more.
- No migration required to start
- Google Workspace, Microsoft 365, WhatsApp Business, Shopify, your ERP. It gets built on top of what already runs.
- Measured or it doesn't count
- Every automation ships with the figure before and the figure after. If it didn't improve, it isn't billed as an improvement.
- The knowledge stays with you
- Your team learns to use the AI; I keep the integration layer. If the engagement ends and the operation stops, the engagement failed.
Proof
Numbers, not adjectives
Everything below is verifiable in the CV, the portfolio, the repositories and Credly.
Verifiable credential
Claude Partner Badge · Claude Code
It certifies the ability to scope, deploy, configure and run Claude Code activations end to end. Eight courses covering the full cycle of a client engagement, with real deliverables — configuration packs, deployment decision docs, security questionnaires, custom commands, skills, hooks and telemetry configs — and a scenario-based capstone assessment.
10years building software for companies
35+projects delivered since 2016
85%of commits AI-assisted, in a real team
5people led as Engineering Lead
The fine print, for your IT team
- 90.7% test coverage on the financial platform
- 443 PRs with a 2.4 h median lead time (DORA Elite band)
- 1 release every 4.6 days, with release engineering built from zero
- 228 endpoints, 60 controllers and 182 database migrations
- 248 of 309 architecture decision records written
- 327 AI skill, agent and rule files in operation
Companies that already trusted
Design, development and architecture projects delivered since 2016. AI consulting is the evolution of this work, not a leap into the dark.
- Sul Finance
- EnXingaPay
- REYSE
- Xolvex
- Renova Espacio
- Inbright
Questions
What gets asked before signing
Is this only for WhatsApp?
No. WhatsApp is the example that lands fastest because almost every Mexican company already uses it, but the work is connecting everything connectable: email, Drive, sheets, calendar, Shopify, your ERP, your CRM, and custom APIs where no off-the-shelf integration exists. The value shows up when processes connect to each other, not when one gets automated on its own.
What is n8n and why does it matter?
It's where the connections live: the layer that moves information between your systems, triggers the processes and enforces the permissions. It can run on your own infrastructure, which means the data doesn't have to leave your control. It's the hard layer of the work and it's the part I build.
Will my data train an AI model?
No. The models operate under terms where your information isn't used for training, and access is limited to the folders and sheets you authorize, one by one. The permissions stay in your company's name, not mine.
What about the data protection law and my privacy notice?
The law in force since March 2025 does not explicitly address AI processing, and the supervisory body changed. In the diagnosis we review what purposes your current privacy notice declares and whether the use you want to make of your historical data fits within them. I'm not your lawyer and I don't replace your legal counsel: I hand them a concrete design and an audit log they can review.
I already have a bot running. Is it worth anything?
It's worth something as a starting point, and usually as a warning. The audit reviews what it can reach today, which permissions are open beyond need, and what isn't being logged. It's the cheapest work on the list and the one that removes the most risk.
Does this replace my team?
No. It replaces the work of typing the same thing three times. What happens in practice is that the team stops entering data and starts serving clients. If your goal is cutting headcount, I'm not the right provider.
Do I have to change my systems?
No. It's built on the tools already running. If something did have to change, it's stated before we start and justified with numbers.
And if the AI gets it wrong?
That's what the person at the expensive points is for. The AI proposes the appointment and the client confirms it; the AI prepares the quote and someone approves it. Where that control point sits is defined with you before building, and every action is logged so what happened can be reconstructed.
What happens if you stop being available?
The access and credentials belong to your company from day one, everything is documented, and your team is trained to run it. The system is designed to survive my departure, not to depend on my availability.
Next step
Tell me what gets typed twice at your company
That's enough to start seeing the network. And if you already have something running, tell me what it can reach. The first conversation is free, and if this doesn't apply to your operation I'll tell you on that call.
Message on WhatsAppA direct reply, no contact form