Alec S.

I build the AI systems a business actually runs on: agents that take repetitive work off your team, a clinic knowledge base used daily since August, and automations that keep running after I hand over. Top Rated, 100% Job Success, 5+ years on Upwork. What I've shipped recently: • A front-desk knowledge base for an independent eye clinic. Training that lived in 15 spreadsheet tabs and one manager's head became an installed web app: 35 protocol pages, a 41-item price grid, in-app editing with revision history, sign-in locked to the clinic's domain. Live since August 2026, used daily. • A time-off and payroll system for a clinic: Paper slips cost 60 to 100 admin hours a year. Now one form on any phone, an Apps Script engine that checks conflicts, updates the calendar and balances, and builds the payroll report every pay period. All 26 employee records migrated. About 7,000 lines of code, 183 automated tests, $0 a month to run. • Vendor-invoice automation for the clinic's bookkeeper: a vision model reads each PDF, files it to the right vendor folder, and waits for her click. 12 of 12 sample invoices correct at about a cent each. Designed to run on Claude via AWS Bedrock inside the clinic's own account, because patient-named documents never belong in a chat window. • Content and outbound engineering for an online education company with a 1M+ audience: Claude Code and Codex agents with MCP connections into the tools the team already used (Wistia, Circle, Customer. io, Google Workspace, Slack). About 10 hours a week saved. • An AI operating system for a business coach: agents for inbox triage, scheduling replies, voice notes to tasks, and a calendar-to-invoice workflow, governed by one plain-English brain document she edits herself. • A curriculum generation pipeline for a language test-prep company that now handles about 80% of the manual work. How I work: I map the process before I build, and you approve the scope first. Where the answer is already known I use plain deterministic code; the model only handles the messy input, and every AI step has a human review gate and an accuracy check before it touches real data. Every system ships with logs you can read, an undo path, plain-language documentation, a video walkthrough, and 30 days of free bug support after launch, so your team owns it and nothing depends on me. Stack: Claude API and Claude Code, MCP servers, AI agents and tool use, RAG, AWS Bedrock, Python, TypeScript and Node, Google Apps Script and Workspace, Supabase and Postgres, Vercel, Playwright, QuickBooks, and the APIs your business already runs on. I co-founded Stephens AI, a Kansas LLC, and my own company runs on the same kind of system: sub-agents on least-privilege tool lists, 50+ reusable skills, scheduled jobs that leave an inspectable log, and a searchable knowledge base I run the whole practice from. Best fit: owner-run businesses and small teams, healthcare practices especially, that know AI should be saving them time and want someone to wire it into how they actually work. Send me the one task your team does by hand every week that you most wish it didn't. I'll reply with a short walkthrough of how I'd automate it, or tell you straight if AI unnecessary.