Software Engineer · Gen AI Solutions Architect
Designing services that stay correct and observable under failure, retries and partial outages.
Translating business requirements into components, contracts and trade-offs that remain understandable over time.
Decoupling services with asynchronous messaging, handling idempotency, retries and event evolution.
Fronting services and models behind API and AI gateways: auth, rate limiting, routing and observability for both APIs and LLM traffic.
Designing agent-to-agent (A2A) workflows, MCP-based tool access and RAG pipelines as components inside larger systems, not as the whole system.
Building and operating cloud-hosted systems on AWS with an eye on reliability and cost.
I'm Majdi, As a Software Engineer I am trained and experienced in designing, constructing, testing and supporting systems.
My main focus is building fast, accessible, and engaging apps that resolve problems and users enjoy.
I have a strong foundation in advanced software development, system design, and project management. My academic journey at INSAT honed my technical expertise and analytical skills, allowing me to tackle complex software challenges.
Currently, I work as a Gen AI Solutions Architect at Sanofi, where I design technology solutions that align with the company's strategic goals in the healthcare industry.
My focus is on API and AI gateways, agent-to-agent (A2A) and MCP-based agent architectures, and RAG pipelines, mostly built on AWS.
I collaborate closely with cross-functional teams to translate business requirements into scalable, efficient and secure systems, ensuring the seamless integration of these technologies.
A few positions I keep coming back to in my writing.
Resilience is a behavior, not a library. Installing a circuit breaker doesn't give you resilience — noticing you're hurting something and stopping does, and you can still get that wrong with the right library installed.
Every hidden cost eventually becomes someone's incident. Patterns like event sourcing or the outbox pattern trade one class of problem (dual writes, lost updates) for another (replay cost, eventual consistency) — worth it, but only if you're honest about the trade before you ship it.
GenAI agents don't need a new design discipline to be trustworthy — treating them as microservices with a new set of capabilities lets you reuse the best practices we already have for reliability, boundaries and observability.
Notes on distributed systems, architecture, event-driven design and AI engineering.
If you’re diving into microservices or looking to scale your systems, you’ve probably heard of <b>Event-Driven Architecture (EDA)</b>. It’s a buzzword for a reason — EDA is a game changer when it comes to building responsive, scalable applications. But what exactly is it, and how does it stack up against more traditional approaches? Let’s break it down.
Every streaming system, every queue, every webhook integration eventually meets a producer faster than its consumer. Backpressure is what decides whether that moment is a graceful slowdown or a cascading outage.
Conway's Law says systems mirror the communication structure of the organizations that build them. Domain-Driven Design says a bounded context should mirror the domain. Most of the time, the domain loses that argument.
Both are called 'events.' Both get published to a broker. That surface-level similarity is exactly why teams keep publishing one where they should be publishing the other, and why the mistake is so hard to notice until a schema change breaks three unrelated services.