I build production LLM systems, the kind that ship to real users, run in regulated environments, and solve problems that don't have Stack Overflow answers yet.
Most recently I spent five years at Tyler Technologies, where I moved from frontend engineering into AI work after winning a company-wide Hackweek competition with an AI-powered data visualization prototype. That prototype became a real product. I then co-founded the internal AI engineering team that built the tooling and frameworks used across the entire company.
AI / ML AWS Bedrock · Azure AI · Claude (Sonnet/Opus) · GPT-4 · Gemini · RAG · pgvector · MCP protocol · MCP server development · prompt engineering · multi-model orchestration
Frontend React · Redux · JavaScript · TypeScript · HTML5 · CSS3
Backend Node.js · Express.js · Python · Bash · Java · Scala
DevOps / Infra AWS (Bedrock, ECR, ECS, S3, CloudWatch, and broader) · Docker · CI/CD pipelines · aws-vault · Claude Code
Databases PostgreSQL · SOQL · pgvector
Compliance FedRAMP · CJIS
Data Assistant A customer-facing AI platform deployed on public government websites that automated data visualization and natural language querying across Tyler's reporting services. Users could ask questions against live datasets and receive contextually appropriate chart suggestions, replacing a fully manual workflow. Built on Azure AI Studio with RAG architecture and pgvector, integrating GPT-4.
AI Foundry Framework A company-wide internal platform enabling engineering teams across Tyler Technologies to design, build, publish and connect MCP servers, AI agents, custom tools, and multi-step workflows. Adopted enterprise-wide. I was responsible for FedRAMP deployment and CJIS compliance, implementing security audit findings and authoring deployment playbooks for regulated environments.
MCP Code Review Pipeline A multi-model code review tool built within one week of the MCP protocol's public release, one of the first production MCP implementations anywhere. Routes git diffs through Claude, GPT, and Gemini in parallel, then aggregates, deduplicates, and prioritizes findings and suggested fixes.
I came to software engineering as a second career after 20 years in hospitality management. I went back to school, completed a bootcamp, taught myself to code, and within three years was winning hackathons and shipping AI systems to production. I'm comfortable solving problems without documentation because I have spent most of my career doing exactly that.
Outside of work I lead at my church, DM long-running Pathfinder campaigns, and collect LEGO.
linkedin.com/in/markwillisford mark.willisford.devwork@gmail.com