AI engineer building production agents and AI-native infrastructure.
Currently DevOps Engineer II at Pax8 and founder of Respondyr, an AI-driven Google Business Profile review response platform. SMBs today, multi-location chains and enterprise next. Four years shipping production infrastructure. Daily Claude Code user. I build internal tools that other engineers actually adopt.
Read: I'm an infrastructure engineer who ships AI products ↗Principles
What I Build
- 01 AI-in-the-loop production operations. KTLO, agentified. Agents sit in the critical path of upgrades, release gating automation, and customer-facing workflows.
- 02 Internal AI tools other engineers choose to use. A Jira/Confluence MCP server I built solo is now part of 20 ICs' daily workflow.
- 03 Multi-service infrastructure on Kubernetes. Go and Python services, pure GitOps via ArgoCD, multi-provider Terraform across AWS and Cloudflare.
Respondyr ↗
Founder & EngineerAI-driven Google Business Profile review response automation. SMBs today, multi-location chains and enterprise next. Multi-service architecture (Go and Python) on AWS EKS, pure GitOps via ArgoCD, multi-provider Terraform across AWS and Cloudflare. The marketing engine runs as its own AI workflow with agent fallback, sending 100,000 personalized emails per month.
How it works
- ▸ Reviews come in. AI responds in your voice, your tone. You set the rules once; it handles the rest.
- ▸ Negative-review gate. SMS and email escalation to the owner before any response goes live to a one- or two-star review.
- ▸ Vertical-aware outputs. HIPAA-conscious phrasing for healthcare; bar-association-conscious phrasing for law firms.
- ▸ Rules engine. Owner sets guardrails (offers, hours, "never say X"). Agent stays within them every response, every time.
Technical Spec
Pax8 ↗
DevOps Engineer IICloud commerce platform powering 50,000+ MSPs (managed service providers) worldwide. Connects IT solution providers to top cloud vendors (Microsoft, Google, AWS, and others) for unified billing, provisioning, and lifecycle management. One of Denver's largest tech employers.
Impact at Scale
Built a Jira/Confluence MCP server in Python (15+ tools including JQL search and full-field updates on any issue or board). Solo project, adopted by 20 ICs daily.
Trust AI agents in critical production paths at Pax8. ~10x engineering-time reduction at equivalent downtime. Family recipes stay in the family.
Co-led zero-downtime Kafka to AWS MSK migration. 2,000 topics across 52 services. Dual-write replication with Helm-driven per-service cutover.
Shipped opt-in GitHub Actions auto-release pipeline. Organic adoption driven by a slick process; eventually mandated org-wide as foundation for AI-agent-driven releases. Hundreds of deploys / week.
Weekly-rebuilt golden image pipeline + Wiz scan integration. Org banned public base images post-rollout. Migrated 400 repositories to non-root containers, 100% patched via scripted rollout.
Planned and executed migration of a toilsome self-managed Debezium pipeline to AWS DMS for CDC replication (monolith RDS to Redshift, ~50 GB/day). Off-loaded internal analytical reads from prod RDS.
Informal technical lead. Turns ambiguous asks into stakeholder-approved plans in about two days and recruits cross-team capacity without formal authority. People-first leadership: recognition over favoritism, performance rewarded, no toxicity.
Role Spec
System Inventory
Infrastructure
Cloud
Languages
Data
Dev Platform
Production AI Patterns
/api-meter Permit service (not a proxy) coordinating outbound rate limits across vendor APIs (Anthropic, MiniMax, GBP, Google Places, Salesforge, Mailforge, Resend). Token buckets in Redis answer "may I call this now?" without touching payloads or credentials. Priority classes preempt low-priority callers in shared buckets. Discord alerts before credits or monthly caps blow up.
/workflows Workflow first; agent when determinism runs out. Respondyr outreach is a deterministic chain: scanner, three-tier email finder, deliverability verifier, LLM analyzer that drafts sample responses to the prospect's own reviews, then Salesforge for send. The moat is the demo, not the send.
/mcp-patterns Three MCP servers in daily use: Jira/Confluence (built solo, 15+ tools, JQL as a first-class search primitive), Salesforge for outbound orchestration, Playwright for intake-form fallback. Tool granularity, idempotent operations, query languages exposed over pre-baked queries.
/agent-ci Trust is established by tests, not by hope. Agent opens a PR, PR triggers tests, tests pass, release is cut, repeat. The opt-in release pipeline I shipped at Pax8 was designed for exactly this loop. Coverage hardened itself as a second-order effect.
Start a conversation
Montrose, CO. Open to remote and frontier-lab on-site.