Full-Stack Architect · AI Systems · Tel Aviv
SystemsthatScale+
My name is Nir Lichtenstein — 
I build AI-driven platforms from first principles — multi-tenant SaaS, RAG pipelines, international SEO infrastructure, complex permission architectures. Twenty years making systems that outlast their requirements.
What I Actually Build
Not features. Systems with deliberate architecture, documented trade-offs, and long-term maintainability — designed from day one.
AI Systems & Pipelines
Multi-stage prompt architectures, RAG engines with internal memory schemas, structured content generation with evaluation loops and schema-constrained outputs.
High-Scale SaaS Architecture
Multi-tenant platforms with complex permission modeling, modular content engines, webhook systems, and scalable data flows.
International SEO Infrastructure
Geo-targeting systems, hreflang architectures, country-specific URL strategies. SEO as a first-class engineering concern.
The Platform That Operates Website Networks
A multi-tenant SaaS platform for managing unlimited websites under a single control plane — organized as Projects → Sites → Markets — with effortless domain swapping, template-driven launches, and API-first delivery.
Beyond a CMS: it includes vector memory, cosine similarity retrieval, and a governed RAG pipeline that builds contextual content and cross-page relationships — without running AI at request time.
“Not a single website — a scalable operating system for brands, domains, and markets.”
Scaling is a way of thinking at the infrastructure level. This system is built for networks. It separates infrastructure, content, and presentation so each tenant can launch new brands through configuration, keep content consistent through governance, and generate new pages using retrieval-powered memory rather than copy-paste.
Each site can run a unique design system or a shared template pack — while consuming the same API-driven core, the same data layer, and the same versioned content artifacts.
PlayGuidePro is a live, production-grade reference implementation built on this platform. The complete infrastructure — including the control plane, CMS layer, vector memory system, and RAG pipeline — can be reviewed in detail during an interview.
Live Systems
Real platforms, real scale. Each a case study in architecture decisions and engineering depth.
EVP Volunteers
Operational control system for volunteer coordination during large-scale events and emergency scenarios. Includes structured team models (Squads), granular RBAC, event-driven workflows, RESTful API integrations, and real-time dashboards for command-level visibility.
PlayGuidePro
International affiliate platform built on a custom multi-tenant website engine. Market-aware routing, hreflang architecture, governed vector memory, and a contract-first RAG pipeline that generates versioned content artifacts with zero AI at request time.
Donor Management
Financial-grade donor operations platform with multi-tier RBAC, transaction integrity, automated email workflows, and webhook-driven integrations for payment processing and reporting.
MyQuiz
Event management and live trivia platform supporting ticket configuration, attendee management, event-level permissions, automated registration flows, and a modular content engine.
Systems Portfolio
A structured technical narrative platform designed to translate complex architectures into executive-level clarity, decision-ready models, and presentation-grade system visibility.
Forex Trading & CRM Platform
Led end-to-end architecture and delivery of a production Forex trading system and client management platform, overseeing system design, transactional modeling, and cross-functional execution.
Problem → Decision
Not what I built — why I built it that way. The trade-offs that matter.
Scaling multi-country content without multiplying infrastructure complexity
Clone the application per market. Country growth becomes a code management problem.
Architected a single codebase with a market configuration layer. Countries exist as governed config objects — routing, operators, bonuses, and content derive from structured tenant definitions instead of forks.
Evolving permission models without creating regression risk
Scatter role checks across API handlers. Adding a role means refactoring dozens of endpoints.
Centralized context-aware RBAC at the router middleware layer. Permissions evaluated per-organization before execution. Role expansion becomes a configuration update — not a refactor.
Operational visibility close to real-time without persistent socket infrastructure
Introduce a dedicated WebSocket server for all dashboards. High infra overhead for low-frequency updates.
Implemented webhook-driven event propagation with optimistic UI updates and targeted cache revalidation. State converges within seconds without maintaining open connections.
Using AI for content generation without introducing runtime unpredictability
Generate content dynamically on request. High latency, inconsistent output, and no audit trail.
Separated generation from delivery. Content is generated through a RAG pipeline, stored as versioned artifacts, and served statically. AI acts as an infrastructure layer — not a runtime dependency.
How I Think
I approach systems as evolving architectures — not isolated features. The goal is always scalability, semantic coherence, and long-term maintainability.
Every decision has trade-offs. Every trade-off should be deliberate. That's what separates engineering from coding.
Architecture mistakes rarely break on day one. They compound silently until change becomes expensive.
A well-defined layer with clear inputs, outputs, and failure modes — not a black box you hope will work.
Schema design, permission modeling, and content architecture that grows without rewrites.
From DB schema to pixel-perfect UI. Fewer handoffs, faster iteration, better coherence.
Let's
build
something.
Available for senior engineering roles, architecture consulting, and AI system design. Based in Israel, working globally.