Skip to content

00 / Night shift · Amman

Demos are easy.Production isn't.

Bashar Ayyash (Yabasha) — AI Systems Architect. Evaluations, harness design and AI security for regulated industries.

Start a project
Scroll — walk the system, desk to deploy

01 / The work surface

Where prompts become pipelines.

Agent routing, retrieval, LLM orchestration — sketched on paper, hardened in TypeScript and PHP, shipped with a trace on every request.

  • 7f0ed1b9agent.router → resolve142 msOK
  • 3527c4b5rag.retrieve → 12 chunks89 msOK
  • 4d0c9952llm.orchestrate → stream1.2 sOK
  • cb4d7cc6seo.answer-engine → publish51 msOK

02 / The architecture

Every block is a running service.

Web frontend, API core, agents, monitoring — wired by streams carrying real traffic. Open a block to read its manifest.

ON CALLModel routing via OpenRouter, retrieval pipelines, and SEO/AEO generation jobs that run when content ships.

03 / The core

Twenty years, compiled.

The record behind it: fintech platforms, supply-chain APIs and AI products answering real queries every day — led, built, kept running.

Years shipping software
20
Queries per day in prod
10K+
Hours to a reply
<24

04 / The interface

This page is the proof.

Everything here exists and runs in production — including this page. The next system on the desk could be yours.

Start a project
Available for consulting

The person behind the work

Who’s Yabasha?

I’m Bashar Ayyash, known online as Yabasha, an AI Systems Architect based in Amman, Jordan. I focus on regulated industries, AI evaluations, harness design and AI security.

yabasha.dev is my official portfolio, where I share my projects, open-source work and writing about building reliable AI and software systems.

About Bashar Ayyash, known as Yabasha

Systems online

LaravelNext.js 16React 19React NativeTypeScriptBunTailwind v4RAG SystemsAI AgentsOpenAI APIPostgreSQLDocker

What I do

Three crafts, one standard.

AI Engineering

Production RAG assistants and tool-using agents grounded in your docs, tickets, and databases — with citations, evals, monitoring, and guardrails. Not demos.

Read the RAG guide

Full-Stack Platforms

Laravel + Next.js systems with auth, payments, queues, dashboards, and audit trails. Built to stay stable under real traffic, audits, and team handovers.

Explore services

Mobile Apps

Native-feel iOS and Android apps with React Native and Expo — offline sync, payment integrations, slow-device performance, and app-store-ready releases.

Mobile service details

Selected work

Built, shipped, running.

All projects
agile-workflow: AI-Native Agile for Teams That Ship

Featured build

agile-workflow: AI-Native Agile for Teams That Ship

A protocol-first skill that turns AI coding assistants into your entire agile operation—sprints, releases, code reviews, and CI/CD in one install.

Developer ToolsOpen Sourceagileai-assistantsgitgithub

Telemetry

20+

Years engineering

50+

Projects shipped

24

Articles written

10K+

AI queries served daily

Selected engagements

Banking · MENA

Four years inside a regulated bank building production RAG and agent systems that answer in Arabic and English, with evals, guardrails and monitoring designed in rather than added afterwards.

4 yrsin production

How we work

Clear scope, weekly proof, no ceremony.

  1. Scope & success metrics

    2–5 days — clarify the journey, constraints, and risks; define KPIs, SLAs, and acceptance criteria.

  2. Architecture & plan

    3–7 days — choose the approach, map milestones, and lock the first deliverable.

  3. Vertical slice

    1–2 weeks — an end-to-end slice (UI → API → DB → observability) validates direction early.

  4. Iterate to full feature set

    2–6+ weeks — weekly increments: features, edge cases, performance, stakeholder feedback.

  5. Harden, launch, improve

    1–2 weeks + ongoing — testing, security review, monitoring, rollout plan, handover docs.

From the blog

Latest writing.

All articles
Sep 15, 2026Web DevelopmentFindable Is Not Chosen — and That Résumé Won't Save YouYour résumé gets you past the ATS filter. That's all 'findable' means — and it's priced at zero. Here's the 4-action sequence job seekers are using to get chosen in 2026.6 min
Aug 4, 2026Web DevelopmentGraph Engineering Is Mostly Airflow With A New Coat Of PaintThe AI world just invented something called 'graph engineering' — and most of it is a rebrand of the DAG schedulers we've run for a decade. But one delta is real, and it's the only thing worth your attention. Here's the take from someone who's built rung 05 before, in Laravel queues, where only the workers changed.10 min
Jul 9, 2026Web DevelopmentThe $18K Ceiling Breaker: Skills That Actually Move Your NumberLearn the seven high‑impact capabilities—incident ownership, systems thinking, AI orchestration, regulatory fluency, migration architecture, cost arbitrage, and decision documentation—that can boost a developer’s pay by $18K in MENA fintech.6 min

Newsletter

Practical AI + full-stack insights for MENA builders. No spam.

Questions

FAQ

Use RAG when your answer must be grounded in changing or proprietary knowledge (docs, tickets, policies) and you need citations. Consider fine-tuning when you want consistent style/format, domain-specific behavior, or tool-selection patterns—and your knowledge is stable. Many real systems use both: RAG for facts, light tuning for behavior.

Open channel

Have a system that needs to survive production?

Get a senior engineer who owns delivery end-to-end — from architecture to production hardening.

Get an AI implementation plan
hello@yabasha.dev
StatusAvailable
Response< 24h
BasedAmman, JO
EngagementsRemote · MENA