Josef Doornink

• —Leading the observability and evaluation architecture for a production .NET-based AI financial assistant, instrumenting agent workflows with OpenTelemetry and integrating New Relic and Langfuse to provide performance, reliability, and quality metrics across a 25+ engineer organization.
• —Architected and operated AKS production environments supporting 30+ microservices, 14K+ daily requests, and a 99.9% uptime SLA for an HRIS product generating $8.3M+ in ARR and 43% YoY growth.
• —Implemented New Relic distributed tracing and service-level monitoring, improving visibility across distributed systems and reducing mean time to resolution by 50%.
• —Automated infrastructure and operational workflows using Python and Go, eliminating 80+ hours of monthly manual work and increasing deployment velocity by 3x.
• —Built an Azure DevOps release orchestration workflow for 15+ microservices across five to seven environments, simplifying version selection, coordinating multi-service deployments, and providing a consistent rollback mechanism.
• —Led Kubernetes capacity planning and scaling strategies that supported a 200% increase in production traffic without compromising platform reliability.
• —Managed 100+ Azure cloud resources through Terraform and implemented Kubernetes security controls supporting SOC 2 compliance.

• —Developed .NET and Angular SaaS applications and helped migrate legacy on-premises products to Azure, contributing to Trimble’s transition toward cloud-based delivery.

• —Built and optimized RESTful APIs for multi-tenant applications serving thousands of users, with a focus on performance, scalability, and tenant isolation.

B2B marketing technology and global fulfillment company

• —Engineered a multi-tenant B2B e-commerce and fulfillment platform using .NET, C#, and SQL Server, integrating third-party SaaS APIs to support customer ordering and fulfillment workflows.
• —Led the migration of a legacy codebase to a greenfield platform, improving application response times by 40% as measured through New Relic APM while increasing scalability and maintainability.

• —Led testing and product development for a multimillion-dollar NIH-funded research program advancing bone-fixation technologies and FDA-regulated medical devices.
• —Designed and executed biomechanical testing that contributed to 10 peer-reviewed publications and two U.S. patents.

• —Evaluated search-result relevance and digital advertising layouts, providing quality insights to distributed web development teams to improve accuracy and user experience.

• —Architected and built an AKS-based ML training and inference platform for an early-stage AI company, establishing the foundational cloud architecture required to move toward production.
• —Designed Azure resource-management patterns for distributed model training, balancing infrastructure cost, workload reliability, and research iteration speed.
• —Built automated training pipelines, validation checks, and reliability safeguards from the ground up, translating evolving research requirements into repeatable infrastructure that allowed the team to experiment faster.

• —Deployed Qwen2.5-7B-Instruct-AWQ with vLLM across Azure A10 GPU nodes, creating a reproducible Kubernetes environment for cloud-based LLM inference.
• —Implemented event-driven autoscaling with KEDA and workload monitoring through Prometheus and Grafana, identifying and resolving scaling flaps caused by an unsuitable autoscaling signal.
• —Published a three-part Dev.to guide covering GPU selection based on model weights and KV-cache requirements, inference deployment, and reliable autoscaling.

• —Built an agentic Kubernetes root-cause analysis tool that uses Gemini to diagnose cluster failures and recommend remediation actions.
• —Added code-enforced safety controls through confidence thresholds, human approval gates, RBAC, and OPA Gatekeeper policies, creating clear boundaries around AI-generated actions.
• —Exposed diagnostic capabilities through an MCP server and published the project on the official MCP Registry, with documented failure modes and confidence-calibration tradeoffs.

• —Built a multimodal semantic video search engine using CLIP embeddings and Qdrant, enabling natural-language queries such as “find a red truck at night.”
• —Instrumented model and system health using Prometheus and Evidently AI to detect performance degradation and embedding drift.
• —Integrated an LLM-based reliability agent that interprets monitoring signals and triggers retraining workflows through GitHub Actions, extending the guardrailed automation approach developed in K8gentS.

• —Built and published a Python CLI that validates MCP servers and scans AI-agent implementations for security vulnerabilities.
• —Added configurable security levels and threshold-based failure conditions, enabling automated security enforcement within CI/CD pipelines.
• —Generated SARIF-compatible results for integration with existing code-scanning and security tools; released publicly on PyPI as Agent-Lint-CLI.

Continuation patent for the far cortical locking bone screw. A bone screw with dual thread profiles enabling flexible engagement to the far cortex only, allowing dynamic fixation that promotes superior fracture healing.

A bone screw including a head, a shaft, and a tip. The shaft includes a first threaded section, a second threaded section, and an unthreaded section positioned between the first and second threaded sections. Designed to enable dynamic fixation.

Original patent formulation for the orthopedic implant enabling far cortical locking via a flexible-engagement thread profile.

Awarded at the American Academy of Orthopaedic Surgeons (AAOS) 2010 Annual Meeting for the landmark exhibit: 'Effects of Construct Stiffness on Healing of Fractures Stabilized With Locking Plates' (Bottlang M., Doornink J., Fitzpatrick DC, Marsh JL, Augat P., von Rechenberg B., Lesser M., Madey SM).