"description":"Production-grade cloud platform rebuilt from scratch on bare-metal Kubernetes.",
"longDescription":"4-node Talos cluster (3 control plane + 1 worker) with OIDC SSO, GitOps (ArgoCD), CI/CD, Kafka, PostgreSQL, S3 storage, GPU LLM inference, and Temporal workflows.",
"stat":"4 nodes, 20+ services, 99.2% uptime",
"highlight":"AWS rebuilt at home—full stack from compute to observability.",
"title":"Poimen: Agent Workflow Orchestration",
"description":"Poimen (Greek: Ποιμήν) means \"shepherd\" — a guide who tends, orchestrates, and reconciles. Poimen is an intelligent orchestration layer that transforms natural language into durable, distributed agent workflows powered by Temporal.",
"longDescription":"An LLM router analyzes user intent, retrieves relevant knowledge from semantic memory, and generates executable workflow specs — enabling agent deployment at scale where any activity can be wired as a composable step in the reconciliation pipeline. Every workflow is durable, retryable, and observable.",
"Go API gateway + Python ML + Next.js frontend + Cloudflare Tunnel zero-trust"
],
"deepDive":{
"label":"Deep Dive: Homelab Architecture →",
"url":"/homelab"
}
"LLM-Powered Workflow Routing: Natural language → executable WorkflowSpec via reasoning model (api.riotpiao.com). Activity Knowledge Base (9 activities) informs the LLM about timeouts, retry policies, and dependencies—intelligent step ordering and error handling strategies.",
"Memory-Augmented Context Retrieval: RetrieveMemoryActivity queries poimen-memory (Rust semantic search service) for relevant skills and lessons before routing—injecting domain knowledge into prompts for context-aware workflow generation.",
"Generic State Machine Executor: RoutingWorkflow executes any JSON workflow spec with JSONPath parameter chaining (${Step1.output.path}), automatic retries for flaky activities, catch blocks for error recovery, and Temporal's durable execution guarantees—every registered activity a composable building block."
]
},
{
"title":"Poimen Memory System",
"description":"Distributed Graph-RAG infrastructure with hierarchical RBAC and wiki-link indexing.",
"description":"The memory layer behind Poimen — a distributed Graph-RAG system that gives the shepherd long-term recall, semantic search, and hierarchical access control over its knowledge base.",
"longDescription":"Three-tier context retrieval pipeline with PageRank-style link scoring, hybrid search fusion (HNSW + BM25), and OIDC-based access control for multi-tenant knowledge graphs.",
"highlight":"Bidirectional wiki-link indexing with RRF fusion + hierarchical RBAC — 50ms signature match tier, graph-boosted hybrid search tier, Obsidian fallback.",
@@ -175,16 +167,24 @@
]
},
{
"title":"Poimen: Agent Workflow Orchestration",
"description":"Temporal-powered orchestration that transforms natural language into durable, scalable workflow executions.",
"longDescription":"LLM router analyzes user intent, retrieves relevant knowledge from semantic memory, and generates executable workflow specs—enabling agent deployment at scale where any activity can be wired as a step in the reconciliation pipeline.",
"description":"From bare metal to a self-healing, GitOps-driven platform — 4-node cluster running 20+ services with full observability, zero-trust networking, and GPU-accelerated AI inference.",
"longDescription":"Built and operate a production-grade Kubernetes platform on Talos Linux — self-healing nodes, declarative GitOps deployments via ArgoCD, SOPS-encrypted secrets, OIDC single sign-on, and end-to-end observability (Prometheus → Grafana → Loki → Tempo). Every change is auditable, every failure auto-recovers, every service is monitored.",
"highlight":"Production-grade platform engineering — not a hobby cluster. Self-healing infrastructure, GitOps-only deployments, full-stack observability, and GPU inference at home.",
"bullets":[
"LLM-Powered Workflow Routing: Natural language → executable WorkflowSpec via reasoning model (api.riotpiao.com). Activity Knowledge Base (9 activities) informs the LLM about timeouts, retry policies, and dependencies—intelligent step ordering and error handling strategies.",
"Memory-Augmented Context Retrieval: RetrieveMemoryActivity queries poimen-memory (Rust semantic search service) for relevant skills and lessons before routing—injecting domain knowledge into prompts for context-aware workflow generation.",
"Generic State Machine Executor: RoutingWorkflow executes any JSON workflow spec with JSONPath parameter chaining (${Step1.output.path}), automatic retries for flaky activities, catch blocks for error recovery, and Temporal's durable execution guarantees—every registered activity a composable building block."
]
"Talos Linux bare-metal cluster — immutable OS, API-driven node management, self-healing on failure",
"ArgoCD + Kustomize + SOPS — zero-touch GitOps: every deploy is a git commit, every secret is encrypted",
"Authentik OIDC SSO + RBAC — single identity across 20+ services, zero-trust access from day one",
"Prometheus + Grafana + Loki + Tempo — full observability stack: metrics, logs, traces, alerts",
"vLLM GPU inference (Qwen3-32B) + KServe — production AI serving with autoscale and traffic splitting",
"Temporal + Kafka + CloudNativePG — durable workflows, event streaming, HA PostgreSQL with pgvector",
"description":"Poimen (\"shepherd\" in Greek) — a self-learning agent platform with two pillars:\n\n🧠 <b>Memory</b>: <b>Rust</b>-based <b>Graph-RAG</b> service. Fixed-window chunking → <b>TEI</b> embeddings → <b>pgvector</b> + bidirectional graph index. Three-tier retrieval via <b>RRF fusion</b>: signature match (<b>50ms</b>), graph-boosted hybrid search, <b>Obsidian</b> fallback. Self-compacting cache evicts stale embeddings and auto-reconciles the graph on every write.\n\n⚡ <b>Workflows</b>: <b>Go</b> + <b>Temporal</b> durable execution engine. LLM router generates <b>WorkflowSpecs</b> from natural language, informed by an <b>Activity Knowledge Base</b> (9 activities with timeout/retry/dependency metadata). <b>JSONPath</b> parameter chaining, catch-block recovery, <b>exactly-once</b> guarantees. <b>Goal-driven</b>: describe intent → platform assembles, executes, and self-heals."
},
{
"company":"riotpiao.com",
"role":"DevOps / SRE / SDE",
"period":"May 2025 — Present",
"description":"Wanted to understand how LLM serving works at scale—so built an entire cloud platform from scratch on bare-metal K8s. 4 machines (1 GPU node, 1 Dell PowerEdge R520, 2 mini-desktops), 3 control planes for distributed consensus. Learned etcd needs low-latency the hard way—powerline adapters spiked 200ms with 30+ pods, killing consensus. Ran ethernet to the garage. Purchased riotpiao.com from Cloudflare with wildcard DNS. Evolved GitOps from pure Terraform to Terraform (Talos config, barely changes) + ArgoCD (CRD-driven observer pattern for K8s resources). Unified IAM with Authentik OIDC, deployed paperless.riotpiao.com for document workflow integration. Now building Poimen—an AI agent system where the right context + small model = successful task inference."
"description":"Designed and operate a <b>production-grade bare-metal Kubernetes platform</b> from scratch — 4 nodes, <b>3 control planes</b>, <b>20+ services</b>, <b>99.2% uptime</b>.\n\n🏗️ <b>Infrastructure</b>: <b>Talos Linux</b> immutable OS, <b>etcd</b> distributed consensus, <b>Longhorn</b> block storage, <b>MinIO</b> S3, <b>CloudNativePG</b> with <b>pgvector</b>. Resolved etcd leader election failures caused by 200ms network latency — re-architected the physical topology for sub-5ms RTT.\n\n🔄 <b>GitOps & CI/CD</b>: Evolved from pure <b>Terraform</b> to <b>Terraform</b> (node provisioning) + <b>ArgoCD</b> (application delivery). <b>Kustomize</b> overlays, <b>SOPS</b>-encrypted secrets, <b>Forgejo CI</b> with DinD runners. Zero manual deployments — every change is a git commit.\n\n🔐 <b>Security & Identity</b>: <b>Authentik OIDC SSO</b> + <b>RBAC</b> across all services. <b>Cloudflare Tunnel</b> zero-trust ingress, <b>cert-manager</b> auto-TLS, no exposed ports.\n\n📊 <b>Observability</b>: <b>Prometheus</b> + <b>Grafana</b> + <b>Loki</b> + <b>Tempo</b> — full metrics, logs, traces, and alerting stack.\n\n🤖 <b>AI/ML Serving</b>: <b>vLLM</b> GPU inference (<b>Qwen3-32B</b>), <b>KServe</b> model orchestration, <b>TEI</b> embeddings, <b>Ollama</b> lightweight models."
},
{
"company":"RBC",
"role":"Lead Software Engineer",
"period":"Nov 2024 — May 2026",
"description":"Learned Terraform and Temporal here—tools fundamental for hosting infra at ease. Tackled two big problems: (1) Flaky deployments—500+ resource state files timing out, 503 errors, lock contention. Fixed by migrating to JFrog Artifactory, splitting plan/apply phases with immutable artifacts, throttling parallelism. Zero state corruption after. (2) Configuration drift—engineers hotfixing in cloud console, code detached from reality. Built nightly cron drift detection with Slack alerts, cut drift visibility from 3 weeks to <24hrs. Key takeaway: Terraform shines when things barely change; for K8s resources that churn, you need ArgoCD."
"description":"Led infrastructure platform engineering for multi-cloud IaC across <b>12 teams</b>.\n\n🏗️ <b>State Migration</b>: Migrated <b>500+ Terraform</b> resource state files from S3 to <b>JFrog Artifactory</b>. Split plan/apply into <b>immutable artifact pipeline</b> with throttled parallelism — eliminated <b>503 timeouts</b>, lock contention, and <b>zero state corruption</b> post-migration.\n\n🔍 <b>Drift Detection</b>: Built nightly <b>cron-based drift detection</b> with <b>Slack</b> alerting — cut configuration drift visibility from <b>3 weeks → <24 hours</b>. Stopped engineers hotfixing in cloud console without updating code.\n\n⚡ <b>Workflow Orchestration</b>: Introduced <b>Temporal</b> for durable workflow execution across provisioning pipelines. Standardized <b>IaC patterns</b> across teams — <b>3× integration velocity</b>.\n\n<b>Stack</b>: <b>Terraform</b>, <b>OpenShift</b>, <b>Docker</b>, <b>Golang</b>, <b>Grafana</b>, <b>Artifactory</b>, <b>Temporal</b>"
"description":"First job out of grad school—learned STAR method and customer obsession at scale. Owned Distributed-Map end-to-end across 57+ regions, sub-100ms P99. Caught a potentially customer-impacting condition field change during code review and coordinated simultaneous frontend+backend deployment—because deployment alignment keeps the business running. Built Redrive Execution for customers to retry from failure point, plus ops tools. Learned backward-compat is critical when frontend consumes latest images. Owned oncall, built CloudWatch dashboards, wrote runbooks."
"description":"Owned <b>Distributed-Map</b> end-to-end for <b>AWS Step Functions</b> — <b>57+ regions</b>, <b>sub-100ms P99</b> latency.\n\n🚀 <b>Feature Ownership</b>: Launched <b>Redrive Execution</b> — retry-from-failure-point for distributed workflows. Designed <b>JSON state input</b> for larger payloads, unlocking new customer use cases. Caught a breaking <b>condition field change</b> in code review — coordinated simultaneous frontend + backend deployment to prevent customer impact.\n\n📊 <b>Operational Excellence</b>: Built <b>CloudWatch</b> dashboards, authored runbooks, owned <b>oncall rotation</b>. Maintained <b>backwardcompatibility</b> across frontend/backend release cycles.\n\n🔧 <b>Distributed Systems</b>: <b>DynamoDB</b> partition design, <b>gRPC</b> service mesh, <b>disaster recovery</b> planning, <b>multi-region</b> replication.\n\n<b>Stack</b>: <b>Java</b>, <b>AWS</b>, <b>DynamoDB</b>, <b>CloudWatch</b>, <b>gRPC</b>, <b>Step Functions</b>"
},
{
"company":"Titus",
"role":"Software Engineer Intern",
"period":"May — Aug 2019",
"description":"Streamlined Personal Data Detection to detect anomaly exit of classified data—achieved 97.8% accuracy. Built fault-tolerant Golang connector—28% p99 improvement over legacy. Re-integrated SmartRegex with CMake & C++ on Linux/Unix—5x faster deployment."
"description":"Data classification security startup — built detection and connector systems.\n\n🔍 <b>Anomaly Detection</b>: Built <b>Personal Data Detection</b> pipeline to flag classified data exfiltration — <b>97.8% accuracy</b>.\n\n⚡ <b>Connector</b>: Fault-tolerant <b>Golang</b> data connector — <b>28% P99 improvement</b> over legacy system.\n\n🔧 <b>Build System</b>: Re-integrated <b>SmartRegex</b> engine with <b>CMake</b> + <b>C++</b> on Linux/Unix — <b>5× faster deployment</b>.\n\n<b>Stack</b>: <b>Golang</b>, <b>C++</b>, <b>CMake</b>, <b>Docker</b>, <b>Linux</b>"
},
{
"company":"NAV Canada",
"role":"Summer Student",
"period":"May — Aug 2018",
"description":"Maintained enterprise web app CFPS in Agile development process. Built Django NOTAMJ polls app to meet iteration goals. Improved deploy stability with Sonar code coverage. Created FWGS weather briefing interface with ReactJS for ATC."
"description":"Air traffic control software — enterprise web applications for flight planning and weather briefing.\n\n✈️ <b>FWGS Weather Briefing</b>: Built <b>ReactJS</b> weather briefing interface used by <b>air traffic controllers</b> for real-time flight weather data.\n\n🗳️ <b>NOTAMJ Polls</b>: Developed <b>Django</b> polling application to meet sprint iteration goals within <b>Agile</b> workflow.\n\n📊 <b>Quality</b>: Improved deploy stability with <b>SonarQube</b> code coverage integration on enterprise <b>CFPS</b> platform.\n\n<b>Stack</b>: <b>ReactJS</b>, <b>Django</b>, <b>SonarQube</b>, <b>Agile</b>"
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.