AI consulting · Granada / Remote
I help companies work out where artificial intelligence actually pays off — and then get it into production, not into a demo nobody uses. Ten-plus years building backend systems at scale, plus my own AI product running in production.
Agents that actually close tickets: they read your knowledge base, look up the real status of an order, and hand over to a person when they cannot answer reliably. Success is measured in cases resolved.
RAG over documentation, contracts, procedures or regulations: your team asks in plain language and gets an answer citing the exact document and section. Without verifiable citations it is not knowledge — it is a well-written guess.
Automating repetitive processes that eat hours today: triage, classification, drafting, reconciliation. Always with a human in the loop at the points where a mistake costs money or trust.
Lead qualification and enrichment, assisted writing in your brand voice, and analysis of sales conversations to surface objections and patterns that today get lost in the CRM.
Turning unstructured documents into data your systems can use: invoices, delivery notes, contracts, scanned or PDF reports. With schema validation and human review whenever the model is unsure.
AI features inside your own software: semantic search, in-context assistants, content generation or summarisation. Built into your architecture, your permission model and your cost controls.
I get to know how the business works, where time is lost and what data actually exists. From there come the use cases with a clear return — and the ones I rule out: I will tell you plainly where AI is not worth it.
Solution architecture, model and provider selection, cost-per-query estimates and an evaluation plan: how we will know, with numbers, that the system answers well before anyone depends on it.
Development integrated into your systems: real authentication, permissions, tracing and deployment. With tests and automated evals in CI, so every prompt or model change is measurable and reversible.
Quality, cost and real adoption metrics from day one. With usage data we tune what fails and decide, on evidence, whether the next use case is worth the investment.
Custom AI agents for business: automate workflows with reliable agents integrated with your systems, with human oversight where it matters.
RAG implementation consultant: reliable answers over your documents, with source citations. Real production experience. From pilot to production.
AI solution architecture: system design, model selection, evals strategy and a technical roadmap before writing code. Grounded decisions, not hype.
Hands-on AI implementation: LLMs, agents and RAG integrated into your stack with tests, evals and observability. From pilot to production.
I am an engineer, not an agency. The person who runs the diagnosis is the same one who writes the code and the one who answers when something breaks in production. That removes the translation layer between whoever sells the project and whoever builds it — which is where AI projects usually die: not for lack of technology, but because nobody owns the technical decisions all the way through.
The track record is verifiable. I am the founder of urbanisti.co, a production RAG system over Spanish urban-planning regulations that answers with citations to the source for architects and municipal technicians. Before that I built PictoEscritura 2.0, a CDTI-recognised adaptive learning platform with a Rails backend and Python ML/NLP services. Behind all of it sits a decade of backend systems and APIs at scale across EdTech, FinTech and LegalTech. I work in Spanish and English, from Granada, with clients anywhere.
The diagnosis is a fixed price, because its scope is defined up front. Implementation is quoted by scope, once we know what has to be built and what it integrates with. I do not give a generic range because it would be made up: an assistant over existing documentation and an automation that touches invoicing and your ERP have nothing in common. When the diagnosis is done you get a concrete number, and you decide whether to continue.
A useful pilot, with real data and people using it, takes weeks. Getting to production depends mostly on integrations: connecting to legacy systems, getting access approved or satisfying internal requirements is usually the critical path. So I work in phases: something usable and measurable early, even if full scope takes longer.
Two things: access to the data or documents the system will work with, and someone from the business who knows the process and can make decisions. You do not need a data team, tidy documentation or any prior AI work — assessing what state your data is in is part of the diagnosis.
It is a design decision, and it gets made at the start. There are providers with agreements that exclude your data from model training; open models can run on your own infrastructure so nothing leaves your perimeter; and personal data can be anonymised before it ever reaches the model. Each option has a cost and a quality level: I explain the trade-off and you decide.
Ruby on Rails and Python are my home ground, and there I work inside your codebase like any other member of the team. With any other stack — Node, Java, .NET, PHP — integration happens over an API: the AI services live separately and your application consumes them, touching your codebase as little as possible.
Tell me about your case and I'll tell you honestly whether AI is the right tool — and how I would implement it.