Hi, I'm Savvas Alexandrou.
Senior Software Architect & Engineer focused on moving past the hype to build dependable automation, distributed backends, and deterministic LLM infrastructure.
Personal blog and technical notes documenting architectural patterns, system reliability, and production engineering lessons.
Focus & Expertise
Pragmatic AI Integration
LLMs & AGENTSMoving from brittle prompt scripts to graph-based state management and self-healing agent workflows. Focused on ensuring probabilistic LLMs respect deterministic business logic.
Lean Automation Architecture
EfficiencyStripping away operational noise by combining intelligent low-code engines (like n8n) with heavy-duty relational backends (PostgreSQL). Lowering token overhead and latency while building systems that maintain themselves.
Let's Connect
Talk through your project
I enjoy talking through system design challenges and the practical realities of running agent workflows in production. If you are building something in this space or exploring an architectural question, let's grab a coffee or set up a video call.
Featured Deep Dives
From Prompt Tennis to Maturing Agents: Why Context Switching Feels Like Cognitive Decline
When an open chat window turns into continuous prompt tennis, the constant context switching drains attention. Building single-purpose, maturing agents replaces endless babysitting with quiet background runs.
Evaluating Prompts When Decimals Cost Thousands: From Chat Sessions to Synthetic Benchmarks
Why testing prompts in a single chat session creates false confidence, and how to build deterministic synthetic benchmarks for models before deploying skills.
Catching the Treasure: Reflections from Google Cloud Summit Builder Connect Poland
Notes from Warsaw on GitOps guardrails for agents, synthetic evaluations, WebMCP, and the shifting role of frontend interfaces.
The Procurement Paradox: When Saving €50 a Month Creates a €30,000 Maintenance Liability
Lengthy procurement reviews push engineering teams to build custom internal tooling, and AI code generation makes the hidden maintenance cost easier to overlook.
Modern Tooling and Minimalism in Early-Stage Teams
How modern tooling enables early-stage teams to launch with massive feature sets, but emphasizes the importance of radical, disciplined simplicity.
Closing the Feedback Loop: Reducing AI Agent Context Bloat Through Automated Self-Retrospectives
How logging execution metrics and running periodic retrospectives enables AI coding agents to detect recurring inefficiencies, catch false greens, and cut prompt context overhead by ~65%.
Recent Technical Publications
- Nov 6, 2026 From Prompt Tennis to Maturing Agents: Why Context Switching Feels Like Cognitive Decline
- Nov 6, 2026 Evaluating Prompts When Decimals Cost Thousands: From Chat Sessions to Synthetic Benchmarks
- Nov 4, 2026 Catching the Treasure: Reflections from Google Cloud Summit Builder Connect Poland
- Oct 21, 2026 The Procurement Paradox: When Saving €50 a Month Creates a €30,000 Maintenance Liability
- Oct 2, 2026 Modern Tooling and Minimalism in Early-Stage Teams
- Oct 1, 2026 Closing the Feedback Loop: Reducing AI Agent Context Bloat Through Automated Self-Retrospectives
- Sep 17, 2026 The Attention Economics of Automation: Architecting Workflows for Cognitive Bandwidth
- Aug 11, 2026 The Psychology of Automation Friction: Why Humans Reject Opaque Systems (and How Architects Fix It)
