yabasha/yabasha

This repository is a GitHub profile README rather than a runnable codebase — a structured introduction to Bashar Ayyash (Yabasha), an AI systems architect in Amman, Jordan, aimed at employers, collaborators, and anyone evaluating his open-source work. It positions him around a specific thesis: twenty years shipping software, four of them inside a regulated bank building production RAG and agent systems with evals, guardrails, and monitoring in both Arabic and English. The unifying theme is the gap between a working demo and a system that survives real users, and every section of the README serves that framing.

The README doubles as an index to six repositories, and the linked work shows consistent technical judgement. composable-ai-stack is a modular stack that is Bun, Turborepo, and Convex-first; arabic-stt-eval stratifies Arabic speech-to-text evaluation by dialect and defines explicit normalisation tiers; gauntlet-loop makes an agent set a quality bar, run builder-and-critic pairs, compare results blind, and iterate until it wins. Domain expertise is also packaged as reusable agent skills — copywrite-skill builds on Schwartz's awareness model, while geo-seo-claude handles citability scoring and AI crawler analysis.

There is nothing to install here; the value is in the curation and the writing it points to. It is worth a look for the under-served Arabic-language AI evaluation angle, the production-first perspective, and touches like an llms.txt published specifically for machine readers.