AS
Production AI & Data Systems Architect

Andreas Sapountzis

Building production AI and data systems that scale

I design and ship AI systems under real constraints: cost, latency, privacy, and reliability. Best for complex engineering challenges and cross-system architecture.

sapountzis.andreas@gmail.com
86%
cost reduction
88%
latency improvement
90%
less alert triage
30%
revenue impact
1M+
devices modeled
100B+
events processed

Worked with

AIREVMentalycEinbliqYumioEuropean Commission

Problem spaces I work in

Complex engineering challenges that require systems thinking and cross-cutting architectural decisions.

AI Systems Architecture

Designing end-to-end LLM systems with real constraints. Agentic workflows, evaluation frameworks, and quality monitoring for production AI.

Data Platform Engineering

Building pipelines at scale from ingestion to serving. Cost control, performance optimization, and operational excellence.

Production Reliability

Rescuing slow, flaky, expensive systems. Bottleneck diagnosis, production-hardening, and operational ergonomics.

Cross-System Thinking

Architecture decisions that ripple through the ecosystem. Platform-level solutions and infrastructure design.

Selected work

Deep dives into production systems. No prototypes — deployed engineering solving expensive problems.

Making Everything Fast and Cheap

The pipeline was slow, expensive, and operationally awkward.

PolarsContainer AppsInfrastructure
Read

Fixing Therapist Note Generation

Mentalyc's AI-generated therapy notes were inconsistent and therapists were rejecting outputs.

LLM opsPrompt engineeringEval harnesses
Read

Validating Patterns at Scale with AI

The detection system discovered many streaming patterns — but the patterns kept multiplying.

LLM agentsRAGPattern validation
Read

Building NPCs That Feel Alive

Yumio needed game characters that could hold natural conversations with voice input, emotional responses, character-specific knowledge, and synchronized animations.

LLM agentsSTT/TTSVector DB
Read

Outsmarting Microsoft's $500/Month Tax

Einbliq's data pipeline uses Azure Durable Functions — long-running tasks that can take 30+ minutes.

AzureKEDACost optimization
Read

Synthesizing "Super Alerts" with Graph Theory

Continuous, real-time monitoring floods operations teams with hundreds of overlapping alerts for a single underlying issue.

Graph TheoryInformation TheoryLLM
Read
View all work

About

Andreas Sapountzis

Andreas Sapountzis

Staff-level Data & AI Systems Engineer

My journey started in neuroscience research, analyzing complex brain activity patterns and dynamic systems. I then moved into streaming analytics at scale, where I learned to handle real-time constraints and massive throughput. Now I focus on LLM operations and agentic systems, solving novel AI problems in production environments.

I have worked across healthcare, gaming, enterprise tools, and scientific research wherever there are hard technical problems worth solving. M.Eng. in Electrical and Computer Engineering from Aristotle University of Thessaloniki. Based remotely, working with teams worldwide.

LLM systemsAgents & RAGML pipelinesData engineeringCloud infrastructure

Engineering philosophy

  • Build for production when reliability matters
  • Explore novel solutions for unsolved problems
  • Context awareness - applying the right approach
  • Cross-system thinking - every decision impacts the broader architecture
  • Observability - measuring what matters

Get in touch

Interested in discussing engineering challenges or just want to connect?

sapountzis.andreas@gmail.comLinkedIn