How do you integrate AI into an existing Laravel app without rewriting everything?

Typically by adding an AI service layer (queues, retries, rate limits), a small adapter for providers, and an audited tool/API boundary. We start with 1–2 high-value workflows, add observability and evals, then expand. The goal is incremental adoption: minimal disruption, measurable outcomes.

What’s your AI architecture review process?

We review goals and constraints first, then map data sources, grounding strategy, tool boundaries, failure modes, and security. Deliverables usually include: a target architecture diagram, risk register, eval plan, and an implementation roadmap with milestones and metrics.

Can you build tool-using agents with approvals and guardrails?

Yes. For higher-risk actions (payments, deletes, emailing, production changes), we design a human-in-the-loop flow: explicit permission prompts, scoped tools, logging/audit trails, and environment separation. The agent can propose actions; execution happens only after approval.

How do you measure AI assistant quality (beyond vibes)?

We define objective metrics: task success rate, groundedness/citation accuracy, latency, cost per task, and error categories. Then we build an eval set (real queries + edge cases) and run regression tests before releases—so quality improves over time instead of drifting.

What are your React Native production quality standards?

Crash-free sessions, performance on low-end devices, clean navigation/state patterns, offline and sync behavior where needed, secure storage, and release discipline (versioning, staged rollouts, monitoring). We treat mobile as a product surface, not just a UI wrapper.

Do you handle multi-tenant security and data isolation?

Yes. We design tenancy boundaries intentionally (schema vs. DB vs. row-level), enforce them at the query layer, add tenant-aware caching, and include audit logs + admin controls. We also review pitfalls like cross-tenant leaks via background jobs or search indexes.

What engagement models do you offer?

Most work fits one of these: (1) fixed-scope delivery for a well-defined feature, (2) monthly retainer for continuous shipping, or (3) architecture + implementation sprint to unblock a team quickly. We’ll recommend the model that best matches uncertainty and timeline.

What timelines should I expect for a first usable release?

For many products, a vertical-slice release is realistic in 1–2 weeks, with a fuller release in 4–8+ weeks depending on integrations, data complexity, and compliance. We aim for early, working increments rather than a long “big bang” delivery.

How do you handle privacy, sensitive data, and compliance constraints?

We minimize data exposure, avoid sending sensitive fields when not required, and add redaction where appropriate. We also choose provider and deployment options based on constraints (region, retention, logging). The system design includes auditability and access controls from day one.

00 / Night shift · Amman

Demos are easy.Production isn't.

Bashar Ayyash — AI engineer and full-stack tech lead. Two decades of Laravel, Next.js and React Native; now RAG systems and AI agents that survive production.

Q: When should I use RAG vs. fine-tuning?

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.

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

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

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

Copywrite Skill

Protocol-first prompts that make your AI write B2B SaaS copy with specificity, proof, and the right CTA.

AICopywritingPrompt EngineeringB2B SaaS

A Father's Gift: Building Wasan's Art Gallery

I built wasansart.com as a gift for my 7-year-old daughter Wasan—a digital gallery where she can showcase her watercolor paintings with friends, teachers, and family. Using Laravel I created a child-friendly, girly-themed website that makes her artistic journey accessible and inviting for both children and parents.

LaravelNext.JSMySQL

GEX: Global & Local Dependency Auditing CLI

GEX is a cross-runtime CLI tool that solves the headache of losing track of JavaScript dependencies when switching Node.js versions or machines

typescriptnodejsbunjs

Telemetry

20+

Years engineering

50+

Projects shipped

Articles written

10K+

AI queries served daily

What clients say

“Bashar architected our AI-powered customer support system that handles 10,000+ queries daily. His understanding of RAG systems and production-grade AI is rare — he doesn't just build demos, he builds systems that work at scale.”

Enterprise Client

VP of Engineering — Banking Sector (MENA)

10K+ queries/day

How we work

Clear scope, weekly proof, no ceremony.

Scope & success metrics
2–5 days — clarify the journey, constraints, and risks; define KPIs, SLAs, and acceptance criteria.
Architecture & plan
3–7 days — choose the approach, map milestones, and lock the first deliverable.
Vertical slice
1–2 weeks — an end-to-end slice (UI → API → DB → observability) validates direction early.
Iterate to full feature set
2–6+ weeks — weekly increments: features, edge cases, performance, stakeholder feedback.
Harden, launch, improve
1–2 weeks + ongoing — testing, security review, monitoring, rollout plan, handover docs.

From the blog

Latest writing.

All articles

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

Jul 9, 2026Web DevelopmentWhy My AI Prompts Are 12 Words Long — And Yours Should Collapse TooIn "Why My AI Prompts Are 12 Words Long — And Yours Should Collapse Too" the author proves that trimming prompts to a concise 12‑word command can dramatically cut latency, cost, and hallucinations for banking AI agents, especially in Arabic‑focused markets.7 min

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