Stylianos Zacharioudakis — ML Engineer & Researcher
Machine Learning Engineer with 4 research papers, 9 deployed production ML systems, and deep expertise in PyTorch, graph neural networks, LLM inference, and production ML pipelines. CS student at NKUA. Open to ML/DL roles in Canada.
Research Papers
- CEGVR (2026) — LLM + Z3 SMT formal verification. 90.6% certified accuracy via conflict-directed search, 50-point lift over baseline. 7,500 eval runs on Qwen3-30B-A3B / A100-80GB. PDF | Code
- DynaDiff-VLBI (2026, submitted to MNRAS) — Benchmark for dynamic radio-interferometric imaging with the Event Horizon Telescope. PDF | Code
- TrustQueryNet (2026) — Trustworthy medical image classification. 83.5% accuracy on HAM10000 with noise-robust ConvNeXt-Tiny and calibrated selective prediction. PDF | Code
- Speculative Decoding Study (2026) — 1.46x speedup with vLLM + EAGLE-3 on A100 80GB. PDF | Code
Production ML Systems (all deployed with live interactive demos)
- Graph Fraud GNN — PyTorch GNN scoring, 14 graph features, 457 req/s, p95=3.3ms
- Realtime Fraud Guard — Streaming cross-channel scoring, ROC-AUC 1.0, Kafka + Redis Streams
- AML Graph Investigator — Graph-native AML, 3K nodes, LightGBM ROC-AUC 0.87, path-level explanations
- NYC Subway Anomaly — Real-time GTFS-RT + online ML (River), 1000+ stations tracked
- CEGVR (LLM+SMT) — 90.6% certified accuracy, interactive results viewer
- DeID Privacy Studio — PHI/PII redaction, 20+ entity types
- DoubleX Ledger — SERIALIZABLE double-entry, idempotent postings, 94% coverage
- LimitForge RLS — 4-algorithm rate limiter, atomic Redis Lua
- EuroSAT DL Benchmark — 7-paradigm comparison (CNN, TL, SimCLR, CLIP, LoRA), 98.37% best accuracy
Technical Skills
ML/DL: PyTorch, GNN, Transformers/HuggingFace, vLLM/EAGLE-3, SimCLR, CLIP, LoRA/PEFT, scikit-learn, LightGBM, River, spaCy, Z3 SMT
Infrastructure: FastAPI, PostgreSQL, Redis, Kafka, gRPC, Docker, Prometheus, Grafana, GitHub Actions CI/CD
Hardware: NVIDIA A100 (80GB), Google Colab, Hetzner VPS
Contact
stelios@stelioszach.com | GitHub | LinkedIn | Download CV (PDF)