I build tools and methodologies for AI-assisted software engineering. Independent researcher. Based in the US, originally from Uruguay.
In 2026 I published Generative Specification — a programming discipline for the stateless reader. The core argument: the dominant failure mode of AI-assisted development is not incorrect code, it is architectural drift. The fix is a discipline that constrains what a stateless reader can derive from the artifacts alone, across sessions that share no persistent context. I built and tested it across six production projects before writing the paper.
→ Paper (Zenodo): doi.org/10.5281/zenodo.19073543 → Methodology hub: genspec.dev → Experiments + quality gates: github.com/jghiringhelli/generative-specification
The methodology, reproducible experiments, and a community quality gate library — 17 gates mapped to the seven GS properties, open for contribution. If you want to challenge a claim, reproduce an experiment, or propose a gate, this is the place.
Production-grade engineering standards for AI coding assistants. Generates tailored instruction files (CLAUDE.md, .cursor/rules/, Copilot instructions) from curated template blocks matched to your stack. Supports Claude, Cursor, Copilot, Windsurf, Cline, and Aider.
npx forgecraft-mcp setup .Graph-powered code intelligence for AI assistants. Builds a knowledge graph of your codebase — imports, calls, class hierarchies — so Claude, Copilot, and Cursor understand how your code actually connects, not just what files contain.
Tech: TypeScript · Node.js · MCP · PostgreSQL · Knowledge Graphs · LLMs