Projects

Building agentic AI and MLOps capabilities, including private, air-gapped LLM infrastructure, for AI-powered data-fusion and investigation platforms used by law enforcement and government agencies.

Turned a manual, local eval harness into a daily automated pipeline with legal-domain evals, full data lineage, and Grafana dashboards for the whole team.

Gave the business visibility into how users respond to AI-generated legal edits, with acceptance tracking, per-tenant cost, and a custom trace export pipeline.

Led the foundation and production deployment of a Text2SQL assistant for the finance department, serving over 1,500 monthly users.

A Chrome extension that shows ZnanyLekarz ratings next to doctors in the LuxMed patient portal, adds booked visits to Google Calendar or Outlook, and opens directions to the clinic.

Cut critical AI agent errors by 90% and kept the product running through an LLM provider outage with error monitoring, smart retries, multi-vendor fallback, and alerting.

Introduced GitOps to the company: split CI from CD, deployed multi-cluster ArgoCD with LDAP login and deny-by-default RBAC, and set the repo structure the team adopted.

Made Jenkins on Kubernetes safe to upgrade and recover with Configuration as Code, multi-level backups, Kaniko builds, and parametrized Bitbucket webhooks.

Built the backend that controlled access to 15+ internally hosted and Bedrock LLMs, which grew into an internal developer platform used daily by 150+ developers.

Built one monitoring view across dev, test, and prod clusters with Prometheus, Thanos, and Grafana, with long-term metrics in S3, custom Python exporters, and alerting rules.

Deployed open-source LLMs with vLLM on EKS with Karpenter autoscaling, AWQ quantization, and a custom Helm chart, exposed to internal users via FastAPI.

M.Sc. research at the Technion: LSTM autoencoders for anomaly detection in real-world force/torque data and reinforcement learning for controller tuning, cutting assembly time by 37%.