Building production AI systems at one of the world's largest modern financial businesses.
Agentic AI Systems Engineer
Production, not demos
Security from day one
Documents what breaks
Agentic AI Systems Engineer
Production, not demos
Security from day one
Documents what breaks
Welcome!
I build AI agents that work in production and I document what breaks along the way.
Most agent projects fail somewhere between the demo and the second user interaction. That gap is where I work: agent architecture, RAG pipelines, MCP tooling, and the failure handling every tutorial skips.
Right now I build production AI systems with the GenAI & Agentic AI team at XM. Before that I designed agentic workflows and RAG tools at Bizeli.
Security isn’t a phase I add before launch. Every tool an agent can call is an attack surface, every permission is a threat model. Before I expose anything, the first question isn’t “does it work?”, it’s “what’s the blast radius if an adversarial prompt gets through?”
I write the process up on my blog “Stratos on Tech”, the architecture decisions, the trade-offs, and the failures. The test before anything goes up: could a tutorial-writer have written this without shipping anything? If yes, it isn’t ready.
Breaking down complex technical topics into clear, actionable steps. First principles, not hand-waving.
Designing AI agents that work in production, with proper guardrails, security, and reliability.
Building and maintaining secure server environments with Proxmox, Docker, and OPNsense. Full control, full ownership.
Documenting what breaks and why — context overload, prompt injection, state that doesn’t survive the second turn. The failures are the useful part.
Building production AI systems at one of the world's largest modern financial businesses.
-Build advanced AI models.
-Create and continuously refine RAG tools.
-Design and implement agentic AI workflows and MCP-based automation solutions for business process optimization.
-Engineered technical solutions to optimize the e-commerce platform.
-Developed comprehensive reporting systems.
-Leveraged data-driven insights.
Monitored technical aspects of the front-end & back-end delivery for projects.
Honors: "Advanced Coding"
Practical training on the architecture and safe implementation of Claude Code.
This program bridges the gap between AI theory and real-world business use.
Pioneering foundational models and strategic vision to lead the generative AI technological revolution.
International Hellenic University - Sindos Campus
Electronics and Network science
Hackathon submission for the AWS & Dynamous AI Hackathon — placed 15th in the competition. An AI agent that analyzes resumes against job descriptions, identifies alignment gaps, and generates optimized content. Handles real resume structures — parsing documents, scoring sections against role requirements, and suggesting targeted improvements grounded in actual job market patterns.
Natural Language Processing, LLM Integration, Document Parsing, Prompt Engineering, Python
A web application built on proven productivity techniques, designed for users who want structure without complexity. Features task management, goal tracking, and workflow patterns — all in a clean, dark-themed interface built for daily use.
Full-Stack Web Development, TypeScript, Modern UI Architecture, Productivity Systems
Hackathon submission for the Elasticsearch AI Hackathon — recognized as a well-crafted, thoroughly documented project by the judges. A second-brain system that uses the Elastic Agent as its reasoning engine and Elasticsearch as its long-term memory — capturing, indexing, and retrieving knowledge so nothing gets lost. Integrated with the Artemis productivity dashboard to prioritize and sequence daily tasks using proven productivity techniques, turning raw information overload into a structured, actionable workflow you can execute each day.
AI Application Architecture, Knowledge Management, TypeScript, Information Retrieval, Data Processing
A personalized AI chatbot embedded on stratoslouvaris.gr that acts as a digital twin. Visitors ask questions about background, skills, projects, and experience — and get accurate answers grounded in a curated knowledge base of 47 entries across 9 categories. Built with a RAG pipeline, vector search, and real-time SSE streaming, self-hosted on personal infrastructure.
RAG Architecture, Vector Search (sqlite-vec), SSE Streaming, FastAPI, Pydantic AI, Self-Hosted Deployment
A first-principles guide to Agent Skills: how they work, how to build one with skill-creator, and how they differ from MCP servers.
What goes wrong when MCP meets real users — context overload, smelly tool descriptions, prompt-injection-as-RCE — and the infrastructure work that prevents it.
Large language models are brilliant but isolated. The Model Context Protocol (MCP) fixes this by standardizing AI integrations. Just as LSP did for IDEs, MCP lets you write a server once using three primitives, Tools, Resources, and Prompts, so any compliant AI assistant can seamlessly use your
Basic RAG works for demos, but production demands more. Explore 11 advanced strategies, from re-ranking and multi-query to agentic and self-reflective RAG, that solve specific failure modes. Learn how to stack these techniques to build accurate, production-ready AI systems.