Shane Waters
Full-Stack Engineer, AI Agents
I build AI systems for people who never wanted software, and I write the code. React and Next.js on the front, an LLM tool layer over RAG and MCP behind it. On a weekday morning about 412 people open one of my desks. The first question a farm client typed into theirs used a county name the way locals say it, and the desk drew a blank, so I fixed that before lunch. Most of the job is like that.
Selected work
The advisory work behind these is at shanewaters.com/advisory.
- The auction board. Halderman sells farmland by the tract at live auction. A bidder can take one tract, a combination, or the whole farm, so someone has to know at every moment which set of bids wins. For years that was a spreadsheet in the back of the room, two or three minutes per new bid while the auctioneer stalled. I sat through a sale before writing anything. The board I built resolves a new combination in a couple of seconds on every screen in the hall, keeps working if the wifi drops, and has 921 tests behind it, most of them hand-built from their old auction sheets. Their own people run it now, with real money in the room.React, a dynamic-programming bid solver, Supabase Realtime, offline fallback
- A clerk's office on an AI desk. The day's work, the institutional knowledge, and an agent that answers in plain language and cites its sources. An elected official with no AI experience ran a full eight-hour workday on it, including a contract review and an RFQ.
- Sophia, the CrimeCon concierge. Six hundred sessions and four hundred speaker profiles behind a chat that would rather say it does not know than make something up. Eight thousand attendees used it. Three hundred forty-three thousand messages across the run-up and the event weekend, and nobody caught it inventing an answer.Next.js, Claude, retrieval grounded in the brand's own content, a refusal guardrail
Experience
- Halderman is a 90-year-old farm management and land brokerage firm, and I am its technology department, part time by design. Identity and security for the whole company, the backup drills that prove a restore works, the vendor contracts, and the runbooks so staff can set up a new hire without me.
- When leadership decides what to adopt next, I am in the room, and then I go build it. It is the same loop I run for Margin clients from the outside.
- Margin builds one AI desk product and shapes it per person from discovery interviews. A farm management firm, a county auditor, a city clerk, and a state-funded nonprofit run on it today, about 12,928 messages a day between them, and it has stayed up 99 percent of the time.
- The AI layer under all of it is mine: streaming chat, a tool layer over MCP, retrieval that scores every claim against its source, and a grader that fails bad output before a person sees it. The nonprofit uses that grader on a $98,500 engagement where their name goes on every number.
- Production code in React, Next.js, and TypeScript on Vercel, Supabase and Postgres underneath. Every build gets reviewed against field notes from whoever is going to use it.
- Started the first true-crime podcast in 2008. There was no simple way to host audio or build a site people could listen on outside iTunes, so I taught myself to code and built one.
- When the work outgrew my tools, I built my own project manager around how my ADHD brain actually works, then grew the show into a six-show network.
- The back office is mine too, a publishing and automation pipeline that moves every episode from recording to release across platforms.
Tools
Build: React, Next.js, TypeScript, Node, Tailwind CSS, shadcn/ui, Supabase, Postgres, Vercel, Playwright, Sentry, n8n, Figma
AI: Agent orchestration, Tool calling, MCP, LangGraph, RAG, Embeddings (pgvector), LLM evals, Claude API, OpenAI API, Gemini API, Claude Code
Education
Master of Science in Computer Science, University of Colorado Boulder (4.0 GPA, all coursework complete)
Graduate Certificate in Artificial Intelligence, University of Colorado Boulder
Bachelor of Science in Business Administration, Purdue University (3.98 GPA)
Certifications
Google Advanced Data Analytics Professional Certificate, Google UX Design Professional Certificate, Front-End Development Libraries (React), Data Analysis with Python.