AI Systems Engineer / Head of AI

POSITION: AI Systems Engineer / Head of AI
TYPE: Full-Time, Remote

ABOUT US

We are a DTC fitness equipment brand selling directly to customers online. We are a lean, distributed team and we have been building a serious AI-driven operation across every department including customer support, operations, marketing, and finance.

We are not looking for someone to introduce us to AI. We already have live, production AI agents running every day. We need someone to take full ownership of what we have built, unify it into one coherent system, and make it significantly better.
This is a senior, strategic role. You will report directly to the founder and work alongside a small team of people who are already using AI daily in their work.

WHAT THIS ROLE IS ABOUT

Right now we have people across different departments each building and maintaining AI systems in their own lane. One person handles reporting automation. Another handles customer support automation. Another handles marketing. Another handles operations and finance. These systems are only partially talking to each other. No single person is responsible for looking across all of it and asking: is this actually the best way to do this?

That is what you are here to do.

Even if we think what we have is the best, something better probably exists. Your job is to find it and lead us toward it.

THE INFRASTRUCTURE YOU ARE WALKING INTO

Everything runs on an on-premise brain computer connected to an always-on Cloudflare Tunnel for external access. There are no cloud servers. Two cloud surfaces sit above it: Firebase Hosting for our dashboards and the Anthropic API for model calls.

We do not use one AI model. We route per job based on cost, speed, and task:

---> Claude Opus (Anthropic) for complex reasoning and primary agent work
---> Claude Sonnet (Anthropic) for CS draft generation, auto-tagging, faster workloads
---> Claude Haiku (Anthropic) for cheap classification and triage jobs
---> Gemini Flash and Pro (Google) for our fitness program generation pipeline, some report narration, and image generation
---> GPT-4o and GPT-5 (OpenAI) for a legacy CS keyword classifier kept on OpenAI for historical reasons
Model selection is intentional. You will be responsible for setting and enforcing model selection standards across all builds, and for maintaining the internal cost-routing system (max_router.py) that routes some API calls through subscription accounts to control cost.

The team accesses AI through the Claude Desktop App via remote desktop on the shared Mac, through Telegram-hosted bots for team-facing interactions, and through Slack for some agent triggers and outputs. We previously used an OpenAI-based system (OpenClaw) built by a technical partner; some legacy CS and ops automations still live there and need to be audited and migrated.

WHAT IS ALREADY BUILT

You are not starting from zero. These are live systems running in production:
---> REPORTING AND ANALYTICS AGENT: Pulls live data from our e-commerce platform, marketing analytics tool, and customer support platform daily. Auto-populates our CEO command dashboard. Runs CS performance reports, sales and marketing reports, cash position summaries, and inbound ticket counts. Triggered via Slack or runs on a morning schedule.
---> SLACK-FACING ASSISTANT: Handles CEO task briefings, inbox scanning, team status updates, and time-off requests with auto-deny logic for policy violations.
---> CS AUTOMATION LAYER: Auto-tags incoming support tickets using a keyword classifier. Generates AI-drafted responses for CS agents. Early-stage at-risk customer flagging that escalates cancel and return signals to priority. Currently limited by API timeout issues at large ticket volumes.
---> FITNESS PROGRAM GENERATOR: Fully automated pipeline. Customer fills out a form, submission triggers an API call to Claude with a large knowledge base system prompt, the AI generates a personalized 10-week training program, and it gets emailed to the customer as a PDF on a randomized morning schedule. Cost is roughly $0.01 per program.
---> CEO DAILY DASHBOARD: Hosted on Firebase. Tabs covering today's tasks, cash position, email metrics, team status, video pipeline, and SaaS spend. Fully AI-populated, no manual data entry.
---> MARKETING AND OPS DASHBOARD: Built using Claude Code. Includes P&L, customer insights pulling from multiple platforms, and a marketing accountability layer.
---> CASH POSITION MODULE: Python script pulling from our accounting platform and cash flow tracker, parsing Slack for payment data via AI.
---> INTERNAL COST ROUTING SYSTEM: Routes API calls between paid API and subscription accounts based on job type to manage costs.
None of these systems are connected to each other. That is the core problem you are solving.

WHAT YOU ARE HERE TO BUILD

UNIFIED AI ARCHITECTURE
Map all existing systems, identify the gaps and redundancies, and build the connective layer so everything functions as one system instead of several disconnected ones.
HOLISTIC BUSINESS AI LAYER
The founder wants to be able to ask any business question in real time and get an answer. Example: what percentage of support tickets about order status lead to a return, and if those customers were moved to priority, what is the projected cash flow impact? Nobody can build that end to end right now. You will.
FIX THE CS AT-RISK SYSTEM
Resolve API timeout issues at large ticket volumes. Train pattern recognition on historical ticket data. D ---------- whether to keep the legacy keyword classifier on GPT or rebuild it on Claude, and justify the decision. Connect the drafting bot to historical data so quality improves over time.
STABILIZE THE SLACK INTEGRATION
The Claude-to-Slack integration was tested but is not in production. Build it properly so tea ---------- mbers can get AI assistance without leaving Slack.
OWN API USAGE GOVERNANCE
Tea ---------- mbers currently hit usage limits mid-workflow with no structure managing it. Build and enforce: model selection standards, usage monitoring, key rotation, cost controls, and maintenance of the cost-routing module.
LEAD TEAM AI TRAINING
Systematically upskill every tea ---------- mber from CS agents to marketing to finance. Run a weekly AI knowledge-share call where primary AI users each teach one new thing per week. The goal is to make every person on the team significantly more effective so we never need to grow headcount just to get more done.
AUDIT AND MIGRATE LEGACY SYSTEMS
Full audit of remaining OpenClaw dependencies. D ---------- what gets rebuilt on Claude, what stays on OpenAI, what gets retired. Document and execute the migration plan.


REQUIRED SKILLS AND EXPERIENCE

TECHNICAL:
---> Deep proficiency in Claude specifically: Claude Projects, Claude API, Claude Code, system prompt architecture, tool use and MCP integration, prompt engineering for structured output and reasoning, context window management at scale, multi-agent orchestration
---> Strong judgment on model selection: knowing when to use Opus vs Sonnet vs Haiku vs Gemini vs GPT, and how to make cost vs quality tradeoffs at production scale
---> Python and scripting: the entire agent stack is Python-based with Google Apps Script integrations; you must be able to read, debug, and extend it
---> REST API integration: connecting external platforms without a full engineering team; our support, e-commerce, analytics, and accounting platforms all need to stay wired into the AI layer
---> Agentic AI architecture: multi-agent setups, memory management, tool chaining, session management, cost routing; you need to build agents that are reliable in production, not just in demos
---> Firebase and cloud hosting: dashboards run on Firebase; you need to understand deployment and the Cloudflare Tunnel setup on the Mac Mini
---> Telegram bot infrastructure: team-facing bots run through Telegram; you need to be able to build on that layer
---> OpenAI and legacy system familiarity: some systems still run on OpenAI; you need to audit and migrate intelligently

LEADERSHIP:
---> Can look at multiple disconnected department-level AI systems and see the unified architecture underneath them
---> Communicates clearly with non-technical teammates across CS, ops, marketing, and finance
---> Pushes back when the current approach is not the best approach; we want and expect this
---> Teaches and elevates the team, not just builds systems for them
---> Works independently across a remote, distributed team without needing close supervision
---> Owns the weekly AI knowledge-share call and is accountable for the team's AI progress over time

FULL TOOLS AND PLATFORMS LIST
AI and Model Layer:
---> Claude (claude.ai, Claude API, Claude Code, Claude Desktop App)
---> OpenAI (GPT-4o, GPT-5)
---> Google Gemini (Flash and Pro)
---> OpenClaw (legacy OpenAI-based agent system)
Infrastructure and DevOps:
---> Mac Mini (on-premise compute)
---> Cloudflare Tunnel
---> Firebase Hosting
---> Telegram (bot interface layer)
---> Python
---> Google Apps Script
---> SSH (remote access)
---> session-commander and max_router.py (internal cost routing tooling)
Business Platforms (all require or have API integrations):
---> Shopify (e-commerce and orders)
---> Richpanel (CS ticketing and automation)
---> Triple Whale (marketing analytics)
---> ShipHero (fulfillment)
---> Flexport (freight and logistics)
---> Xero (accounting)
---> Gusto and Paychex (payroll)
---> Klaviyo (email marketing)
---> Postscript (SMS marketing)
---> KnoCommerce (post-purchase surveys)
---> JudgeMe (product reviews)
Productivity and Collaboration:
---> Slack (team comms, agent triggers, bot outputs)
---> Google Workspace (Sheets, Forms, Drive, Gmail)
---> Asana (project and task management)
---> Claude.ai Projects (knowledge base and system prompt hosting)
Dashboards and Reporting Surfaces:
---> CEO Daily Dashboard (Google Sheets and Firebase)
---> Marketing Dashboard (Google Sheets and Firebase)
---> CS Dashboard (Firebase)
---> Cashflow API (Firebase)

WHAT SUCCESS LOOKS LIKE
By 30 days: Full audit of all existing systems completed. Architecture map delivered. Top priorities identified and approved. Cost routing fully understood and documented.
By 60 days: CS at-risk flagging live and reliable. Slack integration stable in production. Cost governance enforced. Weekly AI call running. Team upskilling underway.
By 90 days: The unified AI layer is live. The founder can query the business holistically. All department systems are connected. Legacy system audit complete with a clear migration plan. The 6-month AI roadmap is on the table.

HOW TO APPLY

Send an email with the subject line: "Head of AI Application" to ----------
Your email must include all five of the following. Incomplete applications will not be reviewed.

1. YOUR RESUME
Attach your updated resume as a PDF or Word document.

2. BRIEF BACKGROUND SUMMARY
In a few sentences, tell us who you are and what you have been doing. What kind of work have you been doing, what type of teams or businesses have you worked with, and what brought you to this kind of role.

3. YOUR AI EXPERIENCE
Give us a specific breakdown of your AI experience. Do not speak in generalities. We want to know:
---> Which AI tools and models you have worked with and at what depth
---> Whether you have built with the Claude API, OpenAI API, Gemini API, or others
---> Whether you have built multi-agent or agentic systems and what they did
---> Any experience with on-premise compute, tunneling, cost routing, or API governance
---> Any experience training or upskilling teams on AI tools and workflows

4. DEMO VIDEO OF YOUR WORK
Submit a screen recording or video walkthrough of an automation or AI project you built yourself. This is the most important part of your application.

We want to see:
---> Something you actually built, not a tutorial you followed
---> How it works end to end, including the prompt logic, data connections, or agent architecture involved
---> The problem it solves and why you built it the way you did
---> Any tradeoffs you made during the build and what you would improve
Upload the video to Google Drive, Loom, YouTube (unlisted), or any shareable link and include it in your email. There is no required length; show us what is necessary to understand what you built.

5. YOUR COMPREHENSIVE 30-DAY PLAN - sent as a pdf or word document
Based on everything described in this job posting, write a detailed plan for what you would do in your first 30 days. This is not a general onboarding plan. We want a specific, technical plan that covers:
---> How you would audit the existing AI systems and what you would look for
---> How you would approach unifying the disconnected systems into one architecture
---> Which optimizations or fixes you would prioritize first and why
---> Any changes you would plan to make to the current model routing, cost structure, or agent architecture
---> How you would approach getting up to speed with the team and the existing codebase

There is no right or wrong answer here. We are evaluating how you think, how specific you get, and whether your plan reflects a genuine understanding of what we have built and what we need.

This role is for someone who has already done this work at this level. Incomplete applications will not be considered. If you are the right person, your application will show it.