AI Operations Engineer — Find & Fix Our Inefficiencies (Long-Term, Full-Time, Philippines)

We run TKO Suites — flexible office space across 18 US states. Private suites, meeting rooms, virtual offices. Real company, profitable, operating since 2007. Our team spends far too much of the day on work a machine should do. Answering the same lead emails, and handling repetitive tasks that could be automated. We're hiring one person in the Philippines to make that stop. Hours — read this first Core hours: 8:00am–1:00pm US Eastern = 8:00pm–1:00am Manila time. Five hours of live overlap with our US team, every working day. The remaining hours are yours to schedule — most people do that work in Manila daytime, before core hours start. Note that Manila runs 12 hours ahead of Eastern most of the year and 13 hours ahead during US winter, so the window shifts by an hour in November and March. We'll hold the Manila time steady and let the ET side move. This is an evening shift, not a graveyard shift — you'll be finished by 1am and your daytime is largely free. Please only apply if this genuinely works for your household long-term. We'd much rather know now than lose you in month four. The most important part of this job: telling us what to build. We want you to learn our operation & tell us how we can become more efficient with AI. We are not going to hand you a task list. Your first job is to learn how this company actually works. Sit with our owner, our management staff, our sales team. Understand how they really spend their week — not how the job is described, how it's actually done. Then come back and tell us what to fix. We expect you to walk into our weekly meeting with proposals nobody asked for, ranked, with your estimate of what each one is worth in hours and dollars saved. You'll get one protected day a week for exactly this — investigating, shadowing, digging into where time goes. Nobody can assign work into that day. A month where you only completed work we requested is a month you underperformed. The reason this role exists is that we know we're wasting time in places we haven't identified. Finding those places is the job. Then you build the fix. Then you teach the team to use it. Half building, half teaching You'll write real code, and you'll also run a weekly session teaching our non-technical US staff to use these tools themselves. If you only want to build, this isn't the right job — our location managers actually using what you make is the entire point. The stack Laravel + Node.js (tkosuites.com, hosted on GitLab), QuickBooks Online, Stripe, Docusign, Google Workspace, Microsoft 365, Google Ads, CallRail. Our CRM lives inside an internal MS Access application over SQL Server, owned by a separate developer you'd work alongside. There's also a separate website developer. **This is not only a stack integration. We also want claude code integrations with all our local desktops as well. And any other reccomendation as well (i.e. Claude Chrome routines, etc). We don't know what we don't know! Where we want to get to These are outcomes, not specifications. How you get there is your call — we want your recommendation, including telling us if we're aiming at the wrong thing. 1. Our CRM becomes AI-accessible. Our lead and tenant data lives in SQL Server behind an internal Access front end. Today nothing else can reach it. We want that to change, safely, so that: Leads get scored against our own conversion history Renewal risk gets flagged before we lose a tenant Inbound leads get enriched and deduplicated automatically A manager can ask a question in plain language instead of running a report 2. Voice handles the calls that don't need a person. Specifically: After-hours inbound — captured and qualified instead of going to voicemail Outbound follow-up on leads that went quiet Tour scheduling, confirmations, and reminders Every call ends up in the CRM with a transcript and a structured summary. We are explicitly not putting AI in front of existing tenants with problems — our staff being genuinely good with people is what our clients rave about, and we're not automating that away. We don't expect you to build speech systems. We expect you to know the landscape, pick well, and integrate it properly. Judgment about when a call must hand off to a human matters more here than any technical skill. 3. Repetitive internal work runs itself. Multi-step workflows with human approval where it matters: email triage and drafting, AR follow-up against QuickBooks Online, per-location reporting, Google Ads spend analysis, lead routing across 18 markets. 4. Our staff can use these tools without you. Internal tools people actually operate, and a team that gets more capable every month. These are our current guesses at priority. Part of your job is telling us which are wrong and what we've missed entirely. How we'll work together on architecture You'll propose the approach. We'd rather hear "here are two ways, here's the tradeoff, here's what I'd do" than have you build the first thing that works. One constraint that isn't negotiable: our Access developer owns the database. You'll have read access to everything. Anything that writes to production CRM data gets designed with them and approved by them first — because business rules live in the Access application, not just in the schema, and writes that bypass it create records our system considers malformed. You propose, they review, then it ships. We're also careful about who holds credentials to production; expect that conversation early. What we need 4+ years shipping production software; comfortable in PHP/Laravel and JavaScript/Node Real experience building with LLM APIs in front of actual users — not demos Built agentic workflows that run unattended — tool use, multi-step execution, retries, and sensible failure behavior when a step goes wrong Solid on REST APIs, OAuth, and webhooks, and connecting systems that were never designed to talk to each other Working knowledge of SQL Server. You don't need to be a DBA — you need to write good queries, understand what you're reading, and have opinions about safe write paths Excellent spoken English. You'll be teaching live, weekly, to Americans who don't code Business judgment. You can look at a process and estimate what fixing it is worth in dollars Initiative. You're comfortable telling the owner that the thing he asked for is less valuable than the thing he didn't Reliable power and internet, with a backup plan. Please tell us what your backup is Based in the Philippines Nice to have: Voice AI platform experience (Retell, Vapi, Bland, ElevenLabs) or any telephony/CallRail integration work. Retrieval systems over business data. Prompt evaluation and regression testing. QuickBooks Online, Stripe, Docusign, or Google Ads APIs. Claude Code, MCP, or similar agentic tooling. BPO or client-facing background. Any teaching or training experience. A note on "machine learning": we're not training custom models and we don't need a data scientist. Where we say ML, we mean practical things — scoring leads from our own historical conversion data, flagging renewal risk, spotting patterns in how our locations perform. Applied, on our data, in service of a decision someone actually makes. Not needed: ML or data science background — we're not training models, we're wiring good ones into a real business. No degree required. How our hiring works — please read, it's unusual Step 1 — 60-minute working session ($30). You interview us about the business, show us one thing you'd build, then teach us to do something useful ourselves, hands-on. The teaching portion is what we're mainly evaluating. Come ready to have us driving the keyboard, not watching you demo. Step 2 — Paid build project ($300, about 2 days) using real sanitized data from our business. Step 3 — Offer, then a 90-day paid trial at full rate before permanent. We pay for every stage, including when we don't move forward. You'll get a yes or no quickly either way. Terms This is an hourly contract, paid weekly against tracked hours — not a fixed-price project and not a salaried role. Approximately 40 hrs/week, ongoing and long-term. Your hourly rate is set by experience. We'd rather pay well for the right person than fill the seat cheaply, and the rate doesn't change based on which tasks you're working on that week. Approved after-hours work is paid at the same rate — we don't ask for unpaid extra hours. Full budget for tools, API credits, and courses. Philippine holidays respected, 13th month consideration included. You'd report to our Operations Manager and sit in the weekly meeting where the owner ranks the backlog — you pitch your own proposals there alongside the department managers. Requests come through one shared, ranked list rather than four managers competing for your time. To apply Start your proposal with "suite 405" so we know you read this. Proposals opening with a generic pitch get archived unread. Then tell us about one thing you automated that saved a real team real time — what the process was before, what you built, and roughly how many hours a week it saved.