Hello — I'm Lou.

I'm David Alejandro Mireles Llamas — Lou for short. I've been building production systems since before it was fashionable to talk about them, and loust.pro is the umbrella under which I do it: a small systems-architecture firm running real infrastructure for real operators, not a portfolio site with mockups.

The work lives in three layers. Below the surface: a VPS substrate hardened from the bottom up, multi-tenant SaaS cores on Next.js + Astro + TypeScript over PostgreSQL with isolated namespaces, and the open research I index under my public ORCID. Between the surface and the operator: agentic AI infrastructure, network transport hardening under hostile transit, server-side attribution pipelines. Above the surface: the products, dashboards, and showcases you can click through.

Most of what I write runs in Go, Rust, or Python — the languages that earn their operational cost. The runtime choices follow the same rule: Kubernetes when the workload justifies the operator cost, containerized systemd-nspawn when it doesn't, distributed control planes when there's actual coordination to do. Observability is grounded in kernel PSI and copy-on-invalid patterns over native memory pools — these are not aspirations, they are the way the production systems stay debuggable at 3am.

I work at the seam where academic depth, operational seriousness, and production code meet. On the academic side I publish under ORCID 0009-0008-4374-2254 — the most recent work is the math foundation for byte-exact retrieval-augmented generation, anchored as a pre-submission paper. The core results — Chernoff-Hoeffding bounds over FNV-1a 128-dim + L2 + cosine, and the O(1) context-compaction analysis that follows from them — are live as Sovereign RAG math gists for peer review. On the engineering side I maintain a multi-tenant SaaS for agencies, an MCP/plugin layer that lets agents touch real infrastructure safely, and a transport-hardening track that survives hostile transit on Mexican carrier paths. The seam is not a metaphor — every public artifact lives at the intersection and is verifiable from its source.

Heritage matters to me. I am from Mérida, Yucatán, and the operational seriousness I bring to infrastructure work is in part inherited from a family line of small-business operators. It is why I optimize for practical cost, latency, throughput, and reliability — not for performative overengineering. Concretely: I hold my own infrastructure to the same kernel PSI / cgroup v2 budgets I document in the self-hosted runner hardening playbook.

Most of my work happens where different operational grammars collide — agency-side brand language, founder-side product language, engineering-side runtime language. Each side optimizes for a different metric and speaks a different dialect of the same production. My job is to translate between them without flattening precision on either side.

Conecto creativos, agencias y tech. The sentence is shorthand for the work: turning a director's intent into a runtime spec, an agency brief into a deployable deliverable, an engineering constraint into a creative option. Concrete artifacts I have shipped at this seam: the CRM/ERP backbone for agency clients, the marketplace module, the agency dashboard with magic-link auth, multi-platform video pipelines, and the orchestration layer that ties them to the substrate. The transport underneath that orchestration is documented separately — see the 5-tier SSH evasion chain for the runtime that keeps an agency brief reachable when the carrier path is hostile.

When I take a brief, I read for the production underneath before I read the surface. The questions I ask are about ownership, success criteria, failure modes, and the path forward when the first attempt does not land — the same questions I ask on an infrastructure engagement, just dressed differently.