Bob Ulrich

Remote

• › Designed and deployed a multi-agent system for 24/7 deployment support automation, built on Temporal for durable execution and Mastra for agent and workflow definition. Seven specialized agents orchestrate parallel investigation with verification-first RCA (live log search and code tracing across pipeline and customer configuration before any diagnostic narrative), confidence-gated live posting, and in-thread human feedback that triggers self-analysis runs. Consumes the company's AWS Bedrock-backed inference platform for the Anthropic Claude family (Opus, Sonnet, Haiku), with per-agent model selection and a deterministic guardrail layer that prevents agents from proposing workarounds to existing operational policies. Handles thousands of investigations daily
• › Designed an evaluation methodology for AI-generated code review at enterprise scale, enabling narrow-scope review skills (security, compliance, performance, maintainability, operational best practices) to be calibrated against thousands of real production pull requests before reaching developers. Retrospective evaluation against historical incidents demonstrated material preventive value across multiple categories of customer-impacting events. Scaling from test team to org-wide deployment
• › Decreased team debugging of operational issues from minutes to seconds by building a collection of reusable Claude Code skills, encapsulating ~9,800 lines of codified operational knowledge (error registries, diagnostic strategies, troubleshooting workflows) previously locked in Slack threads and tribal knowledge
• › Engineered a three-layer token optimization system reducing per-investigation LLM cost by 44-57% (~62K to ~30K tokens). Built a deterministic confidence gate that skips redundant agent calls on ~70% of investigations without quality loss, pipeline-aware config injection that eliminates ~20K tokens of irrelevant domain knowledge per call, and MCP tool result transformers that parse and compress raw log and code search payloads by 10-15x. Implemented per-run token observability with cost attribution, solving the problem that the LLM gateway exposes no usage API
• › Own reliability of Internal Developer Platform supporting 300+ services across commercial and GovCloud (FedRAMP) environments with 99%+ deployment success rate on AWS (EKS, Lambda, Step Functions, DynamoDB, S3)
• › Architected Terraform-based deployment pipeline using AWS Step Functions as a distributed systems orchestrator with child execution cancellation, error propagation, and execution-aware dependency resolution across multi-environment sequential pipelines
• › Fixed critical GovCloud deployment failures in multi-environment pipelines (EKS, Lambda, Step Functions) by implementing execution-aware dependency resolution, preventing false rejections in concurrent runs
• › Led organization-wide GitHub Enterprise migration enabling 300+ services to exit change moratorium constraints
• › Established SLI/SLO dashboard standards with AI-calibrated error budget thresholds using 60-day baseline data, replacing intuition with statistical analysis (P90, P95, mean+3σ) for Prometheus/Grafana monitoring alerts
• › Built interactive observability dashboards with one-click SLI drilldowns to root cause analysis using Grafana, OpenTelemetry, and Splunk, reducing MTTR for platform incidents
• › Implemented pipeline failure classification metrics differentiating platform errors from service-team configuration issues, improving SLI accuracy from a noisy 42% error rate to actionable signal
• › Added error count metric instrumentation for external service dependencies, closing observability gaps across the deployment platform
• › Recurring presenter to executive leadership and the broader engineering organization on AI platform engineering, code review evaluation, and developer productivity. Drove adoption of multi-agent and AI-assisted reliability practices across teams

Phoenix, AZ

• › Promoted to Principal on Substrate Compute Data Plane team managing 10,000+ hypervisors with capacity planning for local storage fleet
• › Eliminated 90% of manual deployment toil by migrating tmux/script-based process to automated Infrastructure as Code (IaC) pipeline
• › Reduced deployment errors to near-zero, cutting rollback incidents by 75%; authored runbooks and blameless post-mortem processes for incident resolution

Seattle, WA

• › Managed fleet of 15,000+ hypervisors on VM Infrastructure Data Plane team supporting Oracle Cloud compute with capacity planning and toil reduction initiatives
• › Reduced on-call escalations by 30% through trend analysis, runbook automation, and targeted toil elimination of repeat incidents
• › Operated large-scale QEMU/KVM infrastructure with iSCSI storage at 99.9%+ availability; led blameless post-mortems driving reliability improvements

Redmond, WA

• › Led Kubernetes migration enabling 50% faster deployments and reducing infrastructure costs by 30%
• › Built CI/CD pipeline cutting release cycle from weeks to hours, increasing deployment frequency 10x
• › Implemented EFK logging stack reducing mean-time-to-detection (MTTD) for production issues by 60%

Redmond, WA

• › Operated one of the world's largest consumer storage services serving hundreds of millions of users
• › Contributed to Azure microservice migration reducing deployment complexity and improving scalability

Redmond, WA

• › Maintained 99.99% SLA for 100M+ monthly active users as key contributor to 24x7 SRE team
• › Reduced catastrophic storage failures by 67% within 1 year through rigorous process improvements
• › Engineered SharePoint update process reducing farm-wide installation time from days to hours (3x faster)
• › Resolved HP miniport driver flaw eliminating storage communication failures across 10,000+ machines
• › Automated certificate lifecycle reducing manual effort by 80% and eliminating expiration incidents

Redmond, WA

• › Executed load balancer migration for 4,000+ servers enabling hardware SSL offloading and 50% faster handshakes
• › Developed automation scripts reducing manual configuration time by 90% and eliminating human error

Portland, OR

• › Led Active Directory and Exchange migration with zero data loss during M&A, maintaining 100% uptime
• › Reduced WAN costs by 25% and improved branch uptime to 99.5% through Cisco ASA upgrades
• › Cut software deployment time by 80% through SCCM implementation across 50+ branch offices

Bellevue, WA

• › Supported 10+ branch offices and 500+ users across Washington with 95%+ ticket resolution rate
• › Improved network performance by 40% through Cisco Catalyst upgrade with VLAN segmentation