--- title: "Carlos Garavito - AI Engineer" description: "AI engineer who designs and operates production AI agents and the stack around them, on Cloudflare, AWS, Azure and GCP, for clients and in his own projects." canonical_url: "https://cgaravito.dev/en" last_updated: "2026-03-21T00:00:00.000Z" locale: "en" --- # Carlos Garavito - AI Engineer > I design and operate AI agents in production and the whole stack around them: runtime, evals, observability, CI and infrastructure on Cloudflare, AWS, Azure and GCP. I have done it for a scale-up and for marketing, agency and software-factory clients, and I keep doing it in my own projects: a Slack agent runtime on Workers, a swarm code review with sandboxed verifiers and a Shopify sales agent. If your product needs an agent that keeps working after the demo, that is what I build. I'm based in Valencia, Spain, and hold an MSc in Cloud Computing and AI from UCM (2026). I'm open to remote roles across the EU, as an employee or B2B contractor. ## Experience ### Vidext - AI Engineer & Full Stack Developer **Full Time** | Feb 2025 - Present | Valencia, Spain - Own the agent runtime on Cloudflare Workflows and Durable Objects, cutting the blocking turn from 20 s to a 2 s acknowledgement. - Built the Braintrust evals and tracing the team releases against, with tool-routing evals as a release gate. - Exposed the agents through MCP servers with organization-scoped OAuth and a bearer API a partner product uses. - Designed the organization sandbox on Cloudflare containers: isolated shells per user, egress denied by default, subagent delegation. - Run DevOps for the product monorepo: agentic CI/CD with per-branch preview stacks on Cloudflare, runners moved to AWS spot after a shadow A/B, and an in-house code-review lane that replaced a paid service, saving about 15,000 EUR a year. - Built the swarm-review workflow the team ships through: coding agents open the PR, several models review it blind, mutation testing gates the merge and the deploy follows. ### Espinosa Consultores - Full Stack Developer **Contract** | 2024 | Spain - Web applications, e-commerce sites and AI-assisted automations for a Valencia marketing consultancy. - Migrated the consultancy's clients to Google Analytics 4 at the Universal Analytics shutdown, keeping campaign reporting uninterrupted. ### GrupoOro - Full Stack Developer **Contract** | 2023 | Spain - Led the development team of a Malaga digital agency, owning code review and mentoring. - Built web and e-commerce projects for its SME clients. ### The Whiteam - Web Developer **Contract** | 2022 | Spain - Angular and TypeScript front-ends on top of Java microservices at a Madrid software factory. ## Projects - LangChain on coding subscriptions - Runs LangChain and LangGraph programs on a Claude Code, Codex or OpenCode subscription, with tool calling and streaming. Python and TypeScript twins published on PyPI and npm; the TypeScript build sits on a zero-dependency Claude Code core. ([GitHub](https://github.com/cgaravitoq/open-langchain)) - Shopify agent on Telegram - Multi-tenant agent that runs any Shopify store from Telegram, whatever the sector, on Workers, D1, R2, Durable Objects and Queues. The production model was chosen with a bench of 18 scenarios across four models, 18/18 at about $0.0015 per turn. - Slack agent runtime - Self-hosted agent runtime for Slack on Workers, Durable Objects, D1 and Workers AI: streamed replies, one durable conversation per thread and a sliding retention that destroys the conversation when it expires. Retention is 7 days for DMs and 15 for channels; the output sanitizer redacts Slack tokens and secret-looking assignments before anything reaches the thread. ([GitHub](https://github.com/cgaravitoq/agentic-slack)) - Swarm code review - Multi-model code review where every reviewer gets a full checkout in a disposable sandbox and can run the real tests, a verifier tries to break each finding, and only what survives is published to the pull request. Two identities inside the container keep the model credential out of the checkout; an eval harness with a sealed answer key reports recall and false-positive rate. ([GitHub](https://github.com/cgaravitoq/swarm-review)) ## Blog - [Browser automation for AI agents with Browser Use CLI](https://cgaravito.dev/en/blog/browser-automation-ai-agents) - How I connected my AI agent to a real browser with active sessions using Browser Use CLI and the Chrome DevTools Protocol, with Brave support. - [Using Ollama Cloud models with OpenClaw at no cost](https://cgaravito.dev/en/blog/free-ai-ollama-cloud-openclaw) - How I configured OpenClaw to use MiniMax, Kimi, and GLM cloud models through a local Ollama server, without a credit card or subscription. - [Multi-Model Code Auditing: The Council Pattern](https://cgaravito.dev/en/blog/council-of-advisors-multi-model-code-auditing) - A multi-model audit of 5,000 lines of TypeScript found 15 unique problems for $1.20. This is how I ran it, merged the results, and checked them. - [Building Multi-Agent Systems with LLMs](https://cgaravito.dev/en/blog/building-multi-agent-systems) - A practical guide to choosing an orchestration pattern, defining agent contracts, and keeping multi-agent systems observable when tasks fail. ## Skills Agent orchestration, Tool use, MCP, Context engineering, Prompt caching, Evals, Model benchmarking, Braintrust, Langfuse, OpenTelemetry, LangGraph, Claude Agent SDK, Vercel AI SDK, TypeScript, Python, SQL, Bash, Cloudflare Workers, D1, R2, Durable Objects, Queues, Workers AI, Azure Container Apps, Azure AI Foundry, Azure AI Search, AWS Lambda, API Gateway, EC2, S3, Cloud Run, Terraform, Alchemy, Docker, GitHub Actions, Bun, Hono, Effect, Astro, FastAPI, Pydantic, Drizzle ## Contact - Email: carlos@cgaravito.dev - Book an intro call: https://calendly.com/cgaravitoq/session-1-1 - GitHub: https://github.com/cgaravitoq - LinkedIn: https://linkedin.com/in/cgaravitodev - X: https://x.com/cgaravitoqp - YouTube: https://youtube.com/@cgaravitoq-ai