28.08.2026 aktualisiert
verifiziert
Premiumkunde
100 % verfügbarAI Platform & Agentic Systems Engineer · Independent Consultant | MLOps, Kubernetes, LLM Agents, HPC
Bologna, Italien
Nur Remote
Ph.D. Electronics, Telecommunications & IT - University of Bologna (DEI), ItalyÜber mich
Independent consultant building production AI infrastructure that is reliable, observable and auditable. I created KubeIntellect, a peer-reviewed Kubernetes operations system that acts behind human approval. I help teams run Kubernetes, ship ML and build safe LLM agents.
Skills
ObservabilityAI AgentsMLflowOpenTelemetryLLMAzure KubernetesHelmPlatform EngineeringMLOpsLangChainAPIsAmazon Web ServicesMicrosoft AzureCloud ComputingContinuous Integration
AI Infrastructure, MLOps & Kubernetes — what I build and run
I take machine-learning and LLM systems into production and keep them reliable, observable and auditable. Ph.D., University of Bologna. Remote from Bologna (CET), in English.
WHAT I HAVE BUILT
ExaMLOps — architect and lead developer of the MLOps platform of the EuroHPC project SEANERGYS, on a European supercomputer: training pipelines (Prefect, Kubeflow), model lifecycle and versioning (MLflow, DVC, MinIO), metric-gated promotion, drift detection, multi-model serving (KServe, Ray Serve, vLLM, OpenShift AI) across Kubernetes and SLURM.
KubeIntellect — an LLM-driven Kubernetes operations system I created, peer-reviewed in the Journal of Grid Computing. It does not just explain kubectl: it gathers live evidence (kubectl, Prometheus, Loki), reasons across logs, metrics, events, RBAC and workloads, and remediates behind a human approval gate on every destructive action. LangGraph supervisor/worker agents on FastAPI and PostgreSQL, tiered RBAC, Langfuse tracing, Terraform-provisioned AKS.
NovaFabric — capture, replay, lineage and audit for AI-agent executions, built on ClickHouse, DuckDB, Apache Iceberg, PostgreSQL and PyArrow, with OpenTelemetry throughout, PII masking (Presidio), cryptographic signing for tamper-evident provenance, and storage and KMS across AWS, Azure and GCP. Relevant to EU AI Act logging and traceability duties.
AOBench — a benchmark I designed for how well LLM agents handle realistic HPC and cloud queries: MCP tooling, SLURM environments, rigorous scoring.
YazSes — free, open-source offline speech-to-text for Linux, macOS and Windows, on PyPI: on-device inference with llama.cpp, ONNX Runtime and Whisper. Nothing leaves the machine by default.
HOW I WORK WITH AI
I use AI coding harnesses (Claude Code, Anthropic and OpenAI APIs, MCP) daily and review what they produce through pull requests. I build the agent systems too: supervisor/worker orchestration, ReAct loops, human-in-the-loop confirmation, MCP servers and clients, A2A, RAG, LLM observability. I coach teams on where this speeds delivery and where oversight is non-negotiable.
PLATFORM — Kubernetes, OpenShift, AKS, Helm, Terraform, Ansible, Docker, NGINX Ingress, GitOps, GitLab CI/CD, GitHub Actions, Linux, Bash. Cluster architecture, RBAC, multi-tenancy, autoscaling, cost tuning, migrations, day-2 ops, incident response (DevOps, SRE, Platform Engineering).
OBSERVABILITY — Prometheus, Grafana, Loki, Tempo, OpenTelemetry, Alertmanager, SLOs, alerting, anomaly detection on telemetry (my published research area), GPU and energy telemetry.
DATA & ML — Python, FastAPI, PostgreSQL, ClickHouse, DuckDB, Iceberg, PyArrow, Redis, ETL/ELT, data quality and lineage, PyTorch, scikit-learn, PySpark, Dask, SLURM, MPI, time-series and graph ML.
Earlier: seven years of enterprise infrastructure. Certified Ethical Hacker.
Erfahrung
Berufserfahrung5 - 10 Jahre
Freelancer seit2016
Sprachen
EnglishverhandlungssicherItalianGrundkenntnissePersianMutterspracheTurkishMuttersprache
Projekthistorie
Independent consultant on production AI and MLOps platforms: Kubernetes/OpenShift, model serving (KServe, vLLM, Ray), observability, reliability, troubleshooting/RCA and AIOps. Current engagement with a European AI/HPC infrastructure client. Open to additional consulting engagements.
Design and operate end-to-end AI/MLOps platforms for HPC and cloud-native environments; led the AI production system of the EU Horizon EuroHPC project SEANERGYS.
ExaMLOps — end-to-end MLOps platform for HPC workload management: auto-discovery training pipelines (Prefect), model lifecycle and registry (MLflow), multi-model inference serving (Ray Serve, MinIO), observability (Prometheus, Grafana, Loki, Tempo), a REST control plane, an operator dashboard with approval gate, a CLI and a natural-language LLM agent.
KubeIntellect — LLM-orchestrated Kubernetes operations system: domain agents for logs, metrics, configuration, security and lifecycle coordinated by a LangGraph supervisor, human-in-the-loop confirmation on every mutating operation. Peer-reviewed (Journal of Grid Computing).
NovaFabric — self-hosted capture, replay, lineage and audit layer for AI-agent executions: runs captured without code changes, sealed, replayed, diffed and exported as signed evidence bundles.
AOBench — trace-driven, role-aware, RBAC-enforced benchmark for LLM agents in HPC operations (SLURM, telemetry, energy APIs).
DECICE — EU Horizon AIOps/digital-twin project: anomaly detection and telemetry-driven scheduling.
Stack: Kubernetes, OpenShift, Helm, Terraform, Docker, Azure, Prometheus/Grafana/Loki/Tempo, MLflow, Prefect, Ray, KServe, vLLM, Python, FastAPI, PyTorch, LangGraph, MCP, SLURM, Linux, CI/CD.
Ph.D. in Electronics, Telecommunications and Information Technology — design, analysis and management of high-performance computing systems (supervisors Prof. Luca Benini and Prof. Andrea Bartolini).
Built the data and machine-learning pipelines behind large-scale HPC operations: continuous collection and analysis of holistic telemetry from production supercomputing systems, anomaly detection and predictive-maintenance models for compute nodes, and power/thermal characterisation for energy-aware workload management.
Delivered the supporting production infrastructure — time-series monitoring and dashboards, distributed data pipelines over multi-terabyte telemetry, SLURM job-trace analysis, and reproducible model training and evaluation — and published the results in peer-reviewed journals and conferences.
Stack: Python, PyTorch, scikit-learn, PySpark, Dask, SLURM, MPI/OpenMP, Prometheus, Grafana, Linux, Docker, Git.
Zertifikate
Claude Code in Action
Anthropic2026
Introduction to Quantum Computing
CINECA2023
Practical Deep Learning for HPC
PRACE2022
Containerization in HPC
CINECA2022
Machine Learning with AMD GPUs & ROCm
HLRS, University of Stuttgart2021
Advanced Parallel Computing
CINECA2021