15+ years designing scalable distributed systems. Now at the intersection of LLM engineering, multi-agent systems, and MCP servers — shipping AI that actually works in production.
I'm Onur Celikaslan, Principal Software Engineer at Eurofins Scientific in Dublin. My background spans life sciences, healthcare, gaming, oil & gas, and edtech — but my current focus is clear: building intelligent systems with LLMs.
I don't just integrate AI as a feature. I think about it as infrastructure — multi-agent orchestration, MCP server design, context-window management, and production-grade reliability. Clean Architecture and DDD aren't optional extras; they're how I think.
I'm also building RepraId and PulseMe — two independent products — while contributing to the .NET/AI engineering community through technical writing on LinkedIn.
An LLM-powered conversational platform built with .NET Clean Architecture, CQRS/MediatR, and Domain-Driven Design. Features subscription and billing workflows, idempotent webhook processing, secure identity management, and production-grade fault tolerance across distributed services.
A Flutter-based mobile application with a full .NET backend. Healthcare domain product, built and delivered solo — mobile app, backend services, and infrastructure. Production-grade reliability in a regulated environment.
Semantic Kernel + Claude API research pipeline in .NET 10. Multi-agent architecture demonstrating real LLM orchestration patterns. Published on GitHub as companion to a LinkedIn article series.
LLM as intelligent middleware replacing RabbitMQ between ASP.NET Core microservices. Demonstrates semantic routing, intent detection, and context-aware message dispatch — a novel take on event-driven architecture.
Principal Engineer on enterprise lab testing software serving pharmaceutical, food, and environmental industries.
Core backend architect for a facial recognition attendance platform deployed across 250+ branches and 10,000+ employees, handling 1k–100k RPS peak. Built on .NET Core microservices with real-time identity validation and iPad-based capture.
Why the patterns we already know — sagas, choreography, circuit breakers — map directly onto LLM agent orchestration. And why treating AI agents as distributed systems is the right mental model.
Read on LinkedIn →A working multi-agent research pipeline using Semantic Kernel and the Anthropic Claude API. Full source on GitHub, architecture walkthrough, and lessons learned from real implementation.
View on GitHub →What happens when your message broker understands intent? Replacing traditional queue-based routing with LLM-driven semantic dispatch between ASP.NET Core microservices.
Read on LinkedIn →Deep dive into tokenization strategy, hybrid retrieval, cross-encoder re-ranking, and RAGAS evaluation for production RAG pipelines. Practical patterns for .NET engineers.
Read on LinkedIn →An ongoing series covering MCP server design, prompt engineering patterns, multi-agent coordination, and production AI integration — written for senior engineers building real systems.
Follow on LinkedIn →Building MCP servers from scratch. Agentic workflows in production. Memory management across LLM turns. Follow on LinkedIn to get notified.
Open to consulting engagements, technical leadership roles, and collaboration on AI engineering projects. If you're building something ambitious with LLMs, distributed systems, or .NET — reach out.