Skip to content
View tylerbcrawford's full-sized avatar

Block or report tylerbcrawford

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
tylerbcrawford/README.md

Tyler Crawford

I build and operate production infrastructure, and direct AI agents to build it faster. Self-taught Linux systems administrator running a 49-service self-hosted environment on Ubuntu, CompTIA Security+ certified, CompTIA Network+ in progress. I spent 15 years in live sound and audio engineering — touring, running FOH, troubleshooting signal chains under pressure, before moving into infrastructure and security; that's where the instinct to stay calm and diagnose comes from.

AI Orchestration & Evaluation

I direct AI agents to build software, and I evaluate those agents for capability and safety.

  • AI Agent Evaluation at Mindrift. I author and calibrate adversarial and capability test cases for LLM-powered agents. The adversarial side is indirect prompt-injection red-teaming and the capability side uses MCPs and a CI/CD pipeline gated by an LLM-as-judge quality review.
  • llm-eval-harness: Open-source LLM-as-judge harness that scores model outputs against a rubric and gates CI on the result. It's a small, runnable version of the measurement work described above, built to run fully offline with recorded fixtures.
  • Subgeneratorr: Open-source subtitle generator that orchestrates Deepgram Nova-3 with your choice of LLM (Claude / GPT / Gemini / Ollama) for keyterm extraction. Built to fill the gaps Bazarr can't cover.

Production Infrastructure

A real, daily-driver environment, secured and automated end to end.

  • Infrastructure Showcase: Reference architecture for the full 49-service stack, covering Docker Compose, nginx reverse proxy, Google OAuth2 on every web UI, wildcard SSL via certbot, fail2ban, and a 5-wave tiered startup that prevents boot-time CPU spikes.
  • Server Monitoring Suite: Observability stack covering container health, resource-threshold alerts, SSL-expiry checks, and log-health auditing, all with Discord alerting and cooldown/dedup logic.
  • Restic Backup System: Modular, encrypted, versioned backups with offsite Google Drive sync, retention pruning, and integrity verification.
  • Boo Bot: Production Discord bot automating the server's community. It rebrands every notification under one identity, suppresses floods, handles requests, and pulls trailers.
  • Homelab Scripts: Service orchestration and automation, including tiered startup, inotify-driven file watchers, and a 6-source API fallback chain for metadata enrichment.

Cybersecurity Projects (UofT, 2024)

  • Penetration Testing: 3-day pen test covering SQL injection, RCE on Struts/Tomcat, Shellshock, and credential cracking.
  • Splunk SIEM: Custom detection rules for brute-force, SQLi, and DoS, with incident-response writeups.
  • Azure Cloud Security: Multi-VM deployment with NSGs, Jump Box architecture, and Ansible automation.
  • IoT Vulnerability Analysis: TP-Link smart bulb security assessment, CVE reproduction, and MITM demonstration.

Certifications

  • CompTIA Security+ (SY0-701): earned 2026
  • CompTIA Network+ (N10-009): in progress (exam summer 2026)

LinkedIn

Pinned Loading

  1. infrastructure-showcase infrastructure-showcase Public

    Self-hosted infrastructure with 49 services — Docker Compose, nginx, OAuth2, automated monitoring and backups on Ubuntu Server

  2. agent-dispatcher agent-dispatcher Public

    Web dashboard for orchestrating headless AI coding agents (Claude, Gemini, Codex)

    TypeScript

  3. subgeneratorr subgeneratorr Public

    Generate and translate subtitles for Plex, Jellyfin, and Emby — Deepgram Nova-3 transcription with keyterm prompting, plus LLM translation into 49 languages (Claude/GPT/Gemini/Ollama).

    Python 7

  4. llm-eval-harness llm-eval-harness Public

    LLM-as-judge evaluation harness — define rubric cases, score model outputs, gate CI. Runs fully offline in replay mode (no API key); EVAL_LIVE=1 hits the real Anthropic API.

    Python

  5. vsi-splunk-siem vsi-splunk-siem Public

    Custom Splunk SIEM — brute-force and failed-activity detection, HTTP POST-volume and non-US traffic alerts, incident response (UofT Cybersecurity)

  6. rekall-penetration-testing rekall-penetration-testing Public

    Full pen test engagement — SQL injection, RCE, Nmap, Metasploit, Burp Suite (UofT Cybersecurity)