Artificial Intelligence.
Applied AI consultant for e-commerce in Spain: RAG, agents, Vertex AI, fine-tuning. Production use cases. Remote from Zaragoza.
Remote work with companies across Spain, based in Zaragoza.
01
What you get
- A filter for real use cases — which problems justify AI and which are better solved with classic software.
- A defensible GenAI architecture: model selection, vector store, RAG, agents and guardrails.
- A working PoC in weeks, not quarters. Reproducible and delivered as code, not as a demo.
- A cost-per-inference model and a governance plan to keep an experiment from turning into an unexpected bill.
02
How I work
-
01DiscoveryA catalog of candidate use cases. Filtered by value, technical feasibility and data availability.
-
02PoCOne to three fast iterations on the selected cases. Objective metrics: accuracy, latency, cost per request.
-
03ProductionFinal architecture: gateway, observability, continuous evaluation, guardrails and rollback. CI/CD for prompts and models.
-
04AdoptionTeam onboarding, operational runbooks and an evolution plan. AI is a living system, not a release.
03
Who it's for
Companies with data and a concrete problem
You know AI fits somewhere in the business but not where, nor how to measure whether it works.
Product teams
You want to embed GenAI into your product without locking into a single vendor or blowing up the cloud bill.
Creative studios and agencies
You need production-grade multimedia generation pipelines with quality control and attribution.
04
Typical use cases
- A RAG assistant over internal corporate documentation, multi-tenant and permission-aware.
- An agent for ticket processing and back-office with tool-use and human handoff.
- A multimodal product-description generation pipeline at scale (e-commerce).
- Fine-tuning an open-weights model for a specialized domain with controlled cost per inference.
05
Frequently asked questions
-
Q01
OpenAI / Gemini / Anthropic or open-weights?Whatever makes sense per case. I always run the exercise of comparing cost, latency and quality against the alternatives — including self-hosted open models if the volume justifies it.
-
Q02
How much does it cost to take a PoC to production?The gap between a pretty PoC and a production system is 3-10×, almost always due to observability, continuous evaluation and governance. I give you that range before we start.
-
Q03
Do you work with sensitive or regulated data?The environment is built precisely for that: self-hosted models or models with data residency in the EU, with no training of third-party models on your information, and on-premise deployment if your compliance requires it. Sectors like banking, healthcare or telecom already have AI setups compatible with those requirements; we nail down the specific fit during discovery.
06