Ai system engineer

AI AGENT SYSTEMS ENGINEER Location: Remote Type: Contract, full-time hours required Duration: Month to Month, maintenance retainer available after handover Compensation: $2000-2,200 monthly USD Start date: Immediate ROLE OVERVIEW We are a fractional COO agency running a portfolio of client brands. Our org chart, position descriptions, SOP library, and tool stack are already built and documented. This role builds the AI system that runs that org chart. One orchestrator agent operating as Chief of Staff, and specialist agents beneath it, each mapped to an existing role and SOP set. The agents must take real actions inside real tools, hand work to each other the way a team does, and report results back to the founder. This is a build role, not an advisory role. You will architect it, build it, ship it, document it, and train the team on it in 30 days. KEY RESPONSIBILITIES System architecture Design the orchestrator and agent hierarchy, the memory and state model, the inter-agent communication layer, and the hosting environment. Build a reusable agent template so specialist agents are produced at speed rather than hand-coded one at a time. Define guardrails, approval gates, and rollback paths before any agent touches a live client account. Agent development Build the orchestrator agent that receives founder directives by voice, text, Slack, or email, decomposes them into tasks, routes each task to the correct specialist agent, tracks status, chases blockers, and escalates. Build specialist agents, each driven by an existing position description and SOP set. Build the handoff logic that moves work down the chain from one agent to the next without manual intervention. Enforce hard separation between client brands so no asset, offer, or voice from one client ever surfaces in another client's work. Integrations and execution Build the action layer that lets agents send and reply to email, send SMS, post and reply in Slack, and create, assign, and close tasks in ClickUp. Build GoHighLevel execution: workflows, funnels, email and SMS campaigns, tags, custom fields, pipelines, and calendars. Build browser-level automation for any tool or function with no usable API. Build read and write access to Google Drive, Docs, and Sheets. Build data pulls from Stripe, Whop, Meta Ads, and TikTok Ads for reporting. Knowledge layer Build the ingestion pipeline for the existing SOP library, brand guides, and position descriptions. Build retrieval so every agent works from current documentation. Build versioning so an SOP update propagates to every agent that depends on it. Voice interface Build a voice layer where the founder speaks to the orchestrator, receives spoken responses, and voice commands dispatch real tasks. Reporting Build the daily digest covering what shipped, what is blocked, and what needs a human decision. Build weekly KPI reporting per client brand. Build real-time exception alerts for agent failures, spend thresholds, and slipped deliverables. Reliability and handover Implement action logging so every agent action is auditable after the fact. Implement error handling, retries, and failure recovery. Deliver full system documentation. Train the founder and internal team to operate and extend the system. Stand up monitoring and a maintenance process before handover. REQUIRED EXPERIENCE Production multi-agent systems that other people depend on daily. LangGraph, CrewAI, AutoGen, OpenAI Agents SDK, or a custom orchestration layer. Strong Python or TypeScript. LLM API development: Claude, OpenAI, tool use and function calling, MCP. Browser automation: Playwright, Puppeteer, Browserbase, Stagehand, or equivalent. REST API integration, webhook handling, OAuth, secrets management, rate limits, retries, and idempotency. Vector databases and retrieval-augmented generation. Translating written SOPs into agent behavior without a technical spec being written for you. Shipping on a compressed timeline with weekly demonstrated milestones. PREFERRED EXPERIENCE GoHighLevel API and agency-level GoHighLevel. ClickUp API. Slack API and bot development. Twilio. Voice: Vapi, Retell, ElevenLabs, or Whisper. Workflow orchestration: n8n, Windmill, or Temporal. Rapid-build environments: Lovable, Cursor, or Replit. Marketing agency, media, or professional services operations. NOT A FIT Automation experience limited to Zapier and Make. Demos and proofs of concept, but nothing shipped that runs unattended. Requires a full technical spec before starting. Cannot work full-time hours for 30 consecutive days with overlap into US Eastern. DELIVERABLES Week 1: Architecture document. Action layer working against GoHighLevel, ClickUp, Slack, email, and SMS. Reusable agent template. One specialist agent reading an SOP and executing an end-to-end task in a live tool with logging. Week 2: Orchestrator live and dispatching. First division built and operating. Inter-agent handoff proven end to end. Brand separation enforced and tested. Week 3: Full agent roster built. GoHighLevel and ClickUp execution hardened against production work. Approval gates and audit logging live. SOP retrieval live across all agents. Week 4: Voice interface live. Daily digest and weekly KPI reporting live. Error handling and exception alerts live. Documentation delivered. Team trained. Monitoring and maintenance process in place. Each milestone is demonstrated live. Payment is released against milestones. HOW TO APPLY Applications missing item 3 will not be reviewed. 1. One paragraph on the most complex agentic system you have shipped. What it did, who used it, and what broke. 2. Links, repos, or a Loom walkthrough of that system running. 3. Your proposed architecture for this build, in bullet form. Address orchestration framework, memory and state model, inter-agent handoff, execution inside tools with no API, guardrails and approval gates, and hosting. State plainly what you believe can and cannot be completed in 30 days. 4. Confirmation of full-time availability for 30 consecutive days, and your rate, hourly or fixed by milestone. 5. Confirmation of availability for a paid trial task before contract award. Selected applicants will complete a paid trial: build a single agent that reads a provided SOP, executes the described action end to end inside GoHighLevel, logs every step, and reports completion to Slack.