Back home

Writing

Build in PublicForward Deployed EngineeringDISAI

FDE Field Notes: Starting an AI-Native Build With a Facilities Audit

New engagement: designing a custom AI-native operating system for a Mexican waste management company with zero digital infrastructure today. Before any architecture gets decided, it starts with a levantamiento - a real facilities walkthrough.

September 14, 2026
Fiscal TechOpen SourceGraph Theory

How Many Hops Are You From an EFOS?

Traditional 69-B monitoring only asks 'is my supplier blacklisted?' But the risk that kills a deduction lives deeper in the network. efos-risk-graph models invoicing as a directed graph, propagates SAT blacklist risk with per-hop attenuation, explains the exact chain, and flags invoicing carousels. Open-core, verified by hand. Originally published in Spanish.

July 17, 2026
LLM OrchestrationCost OptimizationMCP

80% of Requests Never Hit the Expensive Model

How 3-tier routing cuts cost without degrading quality: a Haiku classifier (~100ms, 60 tokens) -> domain specialists -> Sonnet for full orchestration. Routing as architecture, not afterthought.

May 7, 2026
AI EngineeringBuild in PublicDISAI

Building AI-Native Mexican Fiscal Compliance: 7 Months Solo

CFDI 4.0, EFOS/EDOS blacklist monitoring, CSD state management: rule-dense fiscal compliance is the ideal AI domain. What 'AI-native' actually means when engineered, not marketed.

May 5, 2026
AI StrategyMLOpsEngineering

AI Strategy is Not Business Strategy with AI Bolted On

Why AI initiatives that don't tie back to clear business drivers are expensive experiments. What MLOps, PoC scoping, and responsible design actually look like in practice.

March 11, 2026
MCP ServerPerformanceBenchmarking

122,449 Operations Per Minute: Benchmarking a Production MCP Server

How a minimal n8n Railway deployment and a custom MCP server hit 122x the market leader's throughput, with 0% error rate at 1.16ms average latency.

November 14, 2025