Cayetano Soriano Gallego
CTO and hands-on architect with 15+ years in tech, 8+ on AWS and 7+ as CTO. I've led two monolith-to-microservices transformations, and today I build LLM features, MCP servers and agentic engineering workflows that run in production.
hello@cayetano.dev · linkedin.com/in/csorianogallego · github.com/neoshadybeat · cayetano.dev · Madrid, Spain
The open-source agent I built for a small business back office: it finds the client, builds the quote with totals computed in code, not by the model, and waits for a person before approving anything.
See the code on GitHub ↗Not slides: code running in products. Everything below I've built myself over the last year, in client work and in my own products.
Claude generating bilingual listings and SEO copy for a marketplace; LLM symptom triage in a digital-health product; OpenAI and Gemini for chat and speech-to-text in a voice assistant. Tool calling and structured outputs, not free-form chat.
MCP servers that let agents read and act on product data safely, from a back-office admin server to a 7-module domain server (medication, schedule, calls, timeline…). Internal tools built on the Claude Agent SDK.
Personal data stripped before any prompt, feature-flag kill switches and switchable EU endpoints for GDPR; LLM calls covered by tests with a fake model transport; the right model per task to keep cost in check; Sentry monitoring like any other service.
Claude Code across a dozen repos with project rules (CLAUDE.md) enforced in CI, AI agents in CI/CD and automated dependency updates. As CTO, I introduced LLM-assisted workflows into the engineering process.
Clear scope, upfront prices, and you work directly with me, not with a junior team. Remote, or on-site in Madrid. Spanish or English.
Also on request: voice agents, LLMOps, private AI and EU AI Act, AI project rescue and IoT (quote within 24 h). Prices exclude VAT. AI model usage is passed through at cost.
You tell me the problem; I tell you honestly whether AI is the right tool and what it would take.
I review your systems and data, and you get prioritised findings, a roadmap and a fixed quote.
Short iterations with demos, evals and monitoring, then a clean hand-over to your team.
Two legacy monoliths (Clojure, PHP) decomposed into 20+ containerized microservices on AWS ECS, with Terraform-first infrastructure and 99.99% uptime in production.
Built and scaled engineering organizations as CTO at two companies: OKRs, cross-functional squads, on-call culture, and developer experience as a first-class concern.
Vendor and partner contracts (Telefónica), technical due diligence with investors, city-council stakeholders in four cities, budgets and SLAs. I speak engineering and board.
Context → what I did → measurable outcome. The pattern repeats across companies: inherit something fragile, make it boring and reliable, then let the team move fast on top of it.
I'm a CTO who still reads pull requests. Over 15+ years I've gone from web developer to architect to running technology for two companies, and I've kept one rule: own the whole stack, from the Terraform module to the on-call rotation to the roadmap conversation with the board.
I've taken companies from legacy monoliths to cloud-native microservices on AWS, and I've done the unglamorous work that makes that stick: CI/CD, observability, cost control, security, and technical due diligence with investors.
Today I'm focused on AI in production: LLM features, MCP servers, agentic engineering workflows and the guardrails around them (privacy, tests, cost, observability), built with the same discipline as any other backend service. I speak Spanish and English and work equally well in either.
LLM features, MCP servers and agentic workflows shipped in real products, with privacy, tests and observability built in.
8+ years designing and running AWS platforms: containers on ECS, Terraform-first infrastructure, event streaming.
15+ years across PHP, Java/Spring Boot, Python, Scala and TypeScript; from monoliths to polyglot microservices.
Choosing and operating the right store per workload: key-value, search, cache, wide-column and relational.
CI/CD pipelines, immutable images, on-call and observability; reliability as an engineering practice, not a team.
7+ years as CTO: roadmap, hiring, OKRs, vendor contracts, due diligence, budgets and city-council stakeholders.
Computer Systems Engineering
Hackathon award
Earlier: Zend Certified PHP Engineer, MongoDB Certified DBA & Developer, Oracle MySQL Developer.
To save us both time: this is where I add the most value, and the kind of company where I've done it before.
Yes, in Madrid. Everywhere else I work remotely, in Spanish or English, CET hours.
A week of hands-on review: architecture, data, current or planned AI use and cloud costs. You get a written report, a prioritised roadmap and a fixed quote for the next step, with no obligation to take it.
Whichever fits the task and your constraints: Claude, OpenAI, Gemini or open models hosted in the EU or on your own infrastructure. I am not tied to any vendor.
Personal data is stripped or kept out of prompts, I sign an NDA and a GDPR data-processing agreement, and I prefer EU-hosted endpoints.
You do. Everything I build is delivered in your repositories, documented, and handed over to your team.