Petie Clark

A systems thinker, building things that make sense and matter.

VP-level integration engineer at Tradeweb. Founder of Clearline Technology Methods. I make complex infrastructure talk to itself — then I write about what exploded. Raising my family and a flock of chickens on a few acres in Southeast Georgia.

Explore the work Read the writing

0 % AI cost reduction

0 Services routed

0 LIMS deployments

0 T daily volume

/work

Current roles

current 2025 — present

Tradeweb Markets

VP, Integration Engineering

Building, observing, and supporting the infrastructure that powers $3.8 trillion in daily average trading volume. Identity platforms, automation pipelines, cross-functional engineering — the unglamorous stuff that keeps markets moving.

Entra ID Okta CloudOps Automation

current 2021 — present

Clearline Technology Methods

Founder

Built because I was tired of watching small labs get sold $50K software that needed a $120K engineer to maintain. LIMS that labs actually use, infrastructure they can afford, integrations that don't require a CS degree to troubleshoot.

Clients: Tindall Enterprises, CENLA Laboratories

SENAITE Terraform AWS Docker

/builds

Things I've built

Clearline LIMS

Laboratory information management that doesn't need a PhD to configure. SENAITE-based, Terraform-deployed, Docker-containerized. Currently tracking samples across multiple labs with full audit trails and compliance documentation.

View infrastructure →

Observability Platform

Prometheus, Grafana, and Alertmanager wired across the entire stack. From cloud instances to on-prem workloads. Alerts that wake me up for real problems, not disk-space-at-82-percent false alarms.

sso:
  provider: authentik
  apps: 13
  users: 6
  
  integrated:
    - portainer
    - grafana
    - grist
    - lims
    - homepage

Unified Identity Stack

Authentik SSO tying together 13 applications across the infrastructure. One login. One source of truth. Way fewer "I forgot my password" Slack messages.

Operational

$ docker compose ps

NAME STATUS

traefik Up 45d

authentik Up 45d

litellm Up 12d

lims-backend Up 45d

Docker Compose Orchestration

Multi-host Docker compose stacks running Traefik, Authentik, LiteLLM, LIMS, and more. Not Kubernetes — deliberately. Sometimes simple orchestration is the right orchestration.

/ai-lab

Local AI infrastructure

I don't just use AI — I run it. Local inference, custom orchestration, model benchmarking, and the infrastructure to make it all actually useful. This is the stack I've built to understand, test, and deploy large language models.

$ openclaw status

friday: online (main session)

gateway: running

channels: webchat, imessage

skills: 42 loaded

last heartbeat: 4 min ago

Friday (OpenClaw)

A fully autonomous AI assistant running locally on a Mac Studio M3 Ultra with 512GB RAM. Multi-channel (webchat, iMessage), autonomous heartbeats and cron jobs, custom skills for infrastructure management, and persistent memory across sessions. Not a chatbot — a strategic partner with API access to my entire stack.

How I built it →

models:
  - qwen3.5-35b
  - gemma-4-26b
  - glm-4.7-flash
  - nemotron-cascade-30b
  - gpt-oss-20b

backend: ollama
ram: 512GB
quant: Q4_K_M

Ollama Model Lab

Running 20+ local models on bare metal. Testing Qwen, Gemma, GLM, Nemotron, and GPT-OSS variants across different quantization levels. Benchmarking speed, quality, and context window behavior to find the right model for the right job.

Always testing

litellm:
  cache:
    mode: redis
    ttl: 604800
  
  models:
    - claude-sonnet-4-6
    - kimi-k2.6
    - minimax-m2.7
  
  savings:
    cache_hit: 90%

LiteLLM Proxy

Unified API gateway for 10+ model providers with intelligent routing, prompt caching, and cost tracking. Cut API spend by 40% through aggressive caching and model fallback strategies. Redis-backed with 7-day TTL on cached prompts.

Marin OpenWebUI

Self-hosted ChatGPT alternative running on ctm-ai server. Access to local Ollama models and remote APIs through a single interface. The team's default tool when they need to test a prompt or compare model outputs side-by-side.

Operational

/writing

Recent writing

... Loading

Loading recent posts...

Fetching from blog.petieclark.com

Weekly teardowns of infrastructure decisions

No growth hacks. No "10x your productivity." Just what I built, what broke, and what I learned.

Read the blog →