Blog
What we learn building and operating AI systems for companies across México and Latin America.
What agentic coding is, the agentic coding tools I run daily on Claude Code, the end-to-end workflow, and a scorecard against the 2026 literature.
Six Claude Code subagents with a model each, one orchestrator that writes no code, a separate reviewer, and why I use skills instead of agent teams.
How I use Claude Code hooks to block an agent from ending its turn on a red check: exit code 2, a tree fingerprint, per-repo check.sh, and a trace log.
Context engineering for Claude Code: a short CLAUDE.md pointing at on-demand docs, briefs as the spec, skills as slash-only pipelines, and what I omit.
Production case study: Claude driving a 22-tool MCP server on FastAPI/Postgres, durable agent memory, dual-layer guardrails and a self-grading eval loop.
A guide to set up a simple custom Elasticsearch plugin to rescore documents. Create an index, bulk insert documents and debug line by line.
A step by step guide to implement the Deep CNN-Based Blind Image Quality Predictor (DIQA) in Python
Description of the most recent methods for image quality assessment
How a leading real-estate marketplace in the Middle East used the BRISQUE algorithm to quantitatively detect poor-quality listing images and measure their impact on listing views and leads.
Step by Step to configure an eGPU in Ubuntu 18.04 LTS
An introduction to Local Binary Patterns (LBP) for grayscale and rotation invariant texture classification, with a Cython implementation that runs over 400x faster than pure Python.
Step by step guide to code BRISQUE method for automatic image quality assessment