AI/ML Engineer · Athens

Γιάννης
Λαζαρίδης

— John Lazaridis —

AI/ML Engineer · Multi-Agent Systems & LLMOps
Systems Thinker · Technical Mentor · Athens, Greece

Building AI that holds under pressure.Χτίζω AI που αντέχει στην πίεση.
93% Test Coverage
10K+ Daily Requests
12/15 Mentees Hired
20yr Engineering
View Projects Let's Talk
01 / About

Systems Thinking,
Applied to AI

I came to AI through 20 years of building systems that have to work — mechanical engineering, ISO 9001 quality management, experimental research. That discipline carries directly: design for failure modes, not happy paths.

Today I build production AI: multi-agent LLM platforms with persistent RAG memory, zero-cloud document pipelines, and developer tooling that ships with real test coverage and documentation. I also teach — 12 of 15 engineers I've mentored secured AI/ML roles within six months.

I didn't switch careers. I zoomed out. Mechanical engineering taught me how physical systems fail. Quality management taught me how to prevent failure systematically. AI/ML is the third application of the same discipline: building complex systems that stay reliable under real-world conditions.
Mechanical Engineering → ISO 9001 Quality → AI/ML Engineering
02 / Expertise

Core Capabilities

Production-grade skills evidenced through shipped systems, not course completions.

AI / ML
Python Multi-Agent Systems RAG · ChromaDB LLM Orchestration Prompt Engineering Vision LLMs LangGraph Ollama
Production Engineering
pytest · TDD Flask · FastAPI Docker CI/CD · GitHub Actions Architecture Design Documentation Security Hardening
Systems & Reliability
Systems Thinking ISO 9001 FMEA Root Cause Analysis Failure Mode Design
Frontend & Tools
TypeScript · React Streamlit Linux · Neovim Git ComfyUI
03 / Featured Projects

What I Ship

Production systems, open-source tooling, and documented architecture.

Dolores — Multi-Agent AI Platform
PRODUCTION · v0.14.0-beta
Problem
Most AI projects are single-agent demos. Real-world use cases require orchestration across multiple specialized agents with persistent memory, streaming output, and production-quality testing — none of which exists in a single off-the-shelf tool.
Action
Designed a 5-agent CRISP framework (Narrator, Actor, Author, Photographer, Director) over a single Ollama instance with ChromaDB RAG memory, SSE streaming, and ComfyUI integration for image and video generation. Full semantic versioning, changelog, and architecture documentation.
Outcome
93% pytest coverage across 130+ tests. 10K+ daily requests. v0.14.0-beta with documented upgrade path from v0.0.1-alpha.
93% coverage 10K+ req/day 5 specialized agents v0.14.0-beta
Python Flask ChromaDB Ollama SSE pytest Docker
OCRSuite — Zero-Cloud Document Intelligence
PRODUCTION
Problem
Cloud-based document processing sends sensitive data to third parties and requires constant internet access. Existing local solutions lack the accuracy of vision LLMs.
Action
Built a fully local pipeline using Llama 3.2 Vision and DeepSeek-OCR. Extracts text, LaTeX mathematics, Markdown tables, and figures from degraded scans — zero API calls, zero cloud dependencies. Each stage is independently replaceable, fully documented.
Outcome
Shipping with SPECIFICATION.md, AUDIT_REPORT.md, and 5 supporting technical guides alongside the codebase. Demonstrates documentation discipline carried from ISO 9001.
Python Vision LLMs Ollama Typer CLI Security Audit
hebikata — Live Python Learning Platform
DEPLOYED · Apache 2.0
Problem
Informal mentorship doesn't scale. Students need real-time feedback on code, structured progression, and measurable outcomes — without a teacher looking over their shoulder.
Action
Built a kata-based Python learning platform with real-time pytest validation and spaced-repetition progression. Five themed exercise sets (RPG, Hacking, Crypto, and more). Open-source with full governance: CONTRIBUTING.md, CODE_OF_CONDUCT.md, Apache 2.0 license.
Outcome
Deployed live on Streamlit Cloud. Formalizes the curriculum behind an 80% mentorship placement rate into reusable, open-source tooling.
80% placement rate 12/15 mentees hired
Python Streamlit pytest GitHub Actions
slidesh — Dual-Runtime Presentation Compiler
PRODUCTION
Problem
Developers who live in the terminal need to present slides, but existing tools force a choice between web-based (React/Keynote) and terminal-based (asciinema/cowsay). Maintaining both from separate sources is unsustainable.
Action
Designed a TypeScript monorepo that compiles a single Markdown source into two independent runtimes: a React web app and a native ANSI terminal presenter. AST-based architecture with Shiki syntax highlighting and 5 dark themes.
Outcome
One source, two targets. Web and terminal from the same .md file. Published as open-source monorepo with pnpm workspace management.
TypeScript React pnpm AST Shiki
04 / Mentorship

Team Multiplication

80% Placement Rate · 12 of 15 Students Now in AI
24,962+ tutoring archive entries. A live coding platform (hebikata) built to formalize the curriculum. Multi-session professional AI training programs covering agentic AI, RAG, vector databases, and production deployment.
My students ship tested code on day one. Not because I hand them answers — because I teach them how to think about systems. The best model I ever shipped was a curriculum.
— Γιάννης Λαζαρίδης · Mentor since 2025
05 / Writing

Technical Essays

Written in Greek for local practitioners. Systems thinking, AI engineering, and the Greek tech ecosystem.

Browse Articles →

06 / Connect

Open to Senior AI/ML
Engineering Roles

Athens-based or remote. Hybrid preferred.

AVAILABLE · ΑΝΟΙΧΤΟΣ ΓΙΑ ΕΡΓΑΣΙΑ
> github.com/Rumbleaxe > linkedin.com/in/ioannislazaridis > i.d.lazaridis@gmail.com

Support my open-source work:
Every coffee keeps the agents running and the pipelines shipping.