Skip to content
View stelioszach03's full-sized avatar
🏠
Working from home
🏠
Working from home

Achievements

Achievement: Pull Sharkx2Achievement: Pair ExtraordinaireAchievement: YOLOAchievement: Quickdraw

Achievements

Achievement: Pull Sharkx2Achievement: Pair ExtraordinaireAchievement: YOLOAchievement: Quickdraw

Highlights

  • Pro

Block or report stelioszach03

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
stelioszach03/README.md

Stelios Zacharioudakis

Machine Learning Engineer & Researcher

BSc Computer Science · National & Kapodistrian University of Athens

Building end-to-end ML systems with Docker, CI, observability, and live demos — from research to reproducible pipelines.

Portfolio Paper LinkedIn Email


Open to junior ML / MLE roles · 🇨🇦 Canadian work-authorized (open permit — no sponsorship needed) · Athens, Greece → Toronto

💼 Experience

Role Organization Period Impact
Head Engineer Paphos Medical Association Jun 2022 – present AsklepiosMed platform serving 480 doctors · 99.5% uptime over 12 months · deploy cycle 2 days → 15 minutes
Full-Stack Engineer Medihyal Clinic Jan 2025 – present Booking platform, 200+ monthly reservations · Groq LLM integration reduced manual review time by 80%

🧠 ML / DL

⚙️ Engineering

📄 Research — 4 papers, all with code + reproducible pipelines

Paper Status Result Links
🔬 CEGVR — LLM + Z3 SMT Verifiable Reasoning Manuscript · solo author 90.6% certified accuracy (+50 pts vs baseline) · 7 500 eval runs · Qwen3-30B-A3B on A100 📄 PDF · 💻 Code · 🔬 Demo
🔭 DynaDiff-VLBI — Dynamic Radio Interferometric Imaging Manuscript · solo author EHT benchmark · 3D U-Net with test-time optimization 📄 PDF · 💻 Code
🏥 TrustQueryNet — Trustworthy Medical Image Classification Manuscript · solo author 83.5% acc on HAM10000 · calibrated selective prediction · ConvNeXt-Tiny 📄 PDF · 💻 Code
⚡ Speculative Decoding Study — LLM Inference Optimization Manuscript · solo author 1.46× speedup · vLLM + EAGLE-3 · A100 80 GB 📄 PDF · 💻 Code

🚀 Projects & Live Demos — 8 projects, all with live demos + CI pipelines

🔴 Graph Fraud GNN · CI

Hybrid heuristic + PyTorch GNN scoring. 14 graph-context features, 457 req/s, p95 = 3.3 ms.

Live: Simulate transactions → Score → Threshold slider

PyTorch GNN FastAPI Prometheus Grafana

🛡️ Realtime Fraud Guard · CI

Streaming cross-channel scoring. ROC-AUC 1.0 on holdout. Payments + SMS + email.

Live: Pick channel → Safe/Fraud preset → Score event

Kafka gRPC Redis Streams Prometheus

🕵️ AML Graph Investigator · CI

Graph-native AML triage. 3 061 nodes, 51 758 edges, ROC-AUC 0.87. Path-level explanations.

Live: Type node ID → Score + ego metrics + neighbors

LightGBM NetworkX Neo4j React

🚇 NYC Subway Anomaly · CI

Real-time GTFS-RT + online ML (River). 1 000+ stations, 10s auto-refresh.

Live: Click route pills (1, A, N…) → Filter anomaly feed

River TimescaleDB Next.js Mapbox

🧪 CEGVR — LLM+SMT · CI

90.6% certified accuracy via Z3 UNSAT-core feedback. 7 500 eval runs on A100.

Live: Interactive results viewer — click each arm → breakdown

Z3 Solver vLLM Qwen3-30B-A3B A100

🔒 DeID Privacy Studio · CI

PHI/PII redaction. 20+ entity types, per-label mask/hash/redact policies.

Live: Interactive studio with side-by-side redaction

spaCy NER Celery FastAPI

📒 DoubleX Ledger · CI

SERIALIZABLE double-entry ledger. Idempotent postings, FX settlement, 94% coverage.

Live: Post → Replay same key → Balance unchanged

SQLAlchemy PostgreSQL Prometheus Grafana

⚡ LimitForge RLS · CI

4-algorithm rate limiter. Atomic Redis Lua scripts. 90% coverage.

Live: Pick algorithm → Fire 1-25 parallel calls → Watch bars

Redis Lua PostgreSQL FastAPI

🛰️ EuroSAT DL Benchmark

7 paradigms compared: CNN, Transfer Learning, SimCLR, CLIP, LoRA. 98.37% best accuracy.

LoRA: 98.22% with only 0.36% trainable params

PyTorch SimCLR CLIP LoRA EfficientNet

More projects at stelioszach.com — including clinic systems, medical association platforms, and ongoing research.

Every project above has a live demo, CI pipeline, and reproducible setup.

🌐 stelioszach.com · 💼 LinkedIn · 📧 stelios@stelioszach.com

Pinned Loading

  1. colab-speculative-decoding-speed-lab colab-speculative-decoding-speed-lab Public

    Empirical study of speculative decoding on A100 80GB: 1.46x speedup with vLLM + EAGLE-3. Reproducible Colab notebook + LaTeX paper.

    Jupyter Notebook

  2. graph-fraud-command-center graph-fraud-command-center Public

    Real-time graph fraud detection with hybrid heuristic + PyTorch GNN scoring. 457 req/s, p95=3.3ms. Live interactive dashboard.

    Python

  3. TrustQueryNet TrustQueryNet Public

    Trustworthy medical image classification: noise-robust ConvNeXt-Tiny with 83.5% accuracy, calibrated selective prediction, HAM10000 + ISIC 2019.

    Python

  4. llm-smt-verifiable-reasoning llm-smt-verifiable-reasoning Public

    LLM + Z3 SMT formal verification: 90.6% certified accuracy via conflict-directed search. 50-point lift over one-shot. Paper + 7500 eval runs.

    Python