AI Engineer. Production LLM and agent systems in TypeScript.
I build the unglamorous parts that make AI agents trustworthy in production: MCP servers, multi-agent orchestration, observability, and human-in-the-loop gates. Solo-built and shipped a 43-tool MCP server and a 3-tier Claude orchestration layer, backed by 6,000+ tests, 91%+ coverage, and end-to-end Langfuse tracing. Open to AI engineering roles.
📍 México (UTC-6, EST-aligned) · 🌐 tripl3.dev · ✉️ hola@tripl3.dev · LinkedIn
SAT-MCP. Production MCP server for Mexican tax compliance (CFDI 4.0). Full MCP primitive set plus 8 MCP-UI mini-apps, SSE and HTTP Streamable transport, multi-tenant CSD management, 5 PAC providers behind circuit breakers, and EFOS/EDOS blacklist monitoring. 6,000+ tests in strict TypeScript. Private commercial product; happy to walk through the architecture and code in an interview.
DISAI-Conta. AI-native fiscal platform built on SAT-MCP, live in private beta at disai.mx. A Haiku router classifies 10 fiscal domains in about 100ms, Sonnet domain agents run native tool_use loops with self-correction, and an Expert Registry injects SAT catalog resources into context before the first call, removing an entire class of hallucinated catalog codes without RAG overhead. Langfuse traces everything; a HITL dashboard gates irreversible operations. (Next.js 16, Anthropic SDK, SSE streaming)
Crypto/TradFi Analytics Terminal. Real-time dashboard over dual WebSockets: Node.js backend, Redis caching, circuit breakers, and graceful degradation. (Live demo Archived)
n8n Freelancer Starter. One-click Railway template for production-configured self-hosted n8n. Replaces a $20 to $30 per month Zapier/Make dependency for small teams.
I write about what I actually build: the design decisions, and what breaks in production.
- 122,449 Operations Per Minute: Benchmarking a Production MCP Server
- 80% of Requests Never Hit the Expensive Model: 3-Tier Routing as Architecture
- Building AI-Native Mexican Fiscal Compliance, 7 Months Solo
- AI Strategy Is Not Business Strategy with AI Bolted On
AI systems: Anthropic Claude API (tool_use, streaming, multi-turn) · MCP (full primitive set, server and client) · multi-agent orchestration · Langfuse tracing and evals · HITL approval patterns
Backend: TypeScript / Node.js · PostgreSQL (RLS, pgvector) · SQLite · Redis · Zod · Docker · Railway · CI/CD
Frontend: Next.js · React · Tailwind · SSE streaming UIs
Also: n8n (self-hosted, multi-client) · Python (growing: data pipelines, verification scripts)
- Model Context Protocol: Advanced Topics, Anthropic
- Claude Code in Action, Anthropic
- Building Scalable Agentic Systems, DataCamp
- n8n Self-Hosted for Enterprises, n8n
Full credential history on LinkedIn.
Before AI engineering: 13 years running a tattoo studio and working as a graphic designer, then two years writing deep technical research on protocol architecture and market microstructure (research archive). Precision, pattern recognition, and reading unfamiliar systems fast. The same rigor now goes into AI infrastructure.
Reliability and observability · Evaluation over assumption · Pragmatism over hype