I'm Keyshawn Rollins — founder of Twin City Automation. I design multi-agent AI systems and end-to-end automations that handle the work small businesses can't afford to hire for: content production, compliance auditing, customer outreach, and operational reporting. Built in n8n, Zapier, and a deeply layered AI stack.
A short look at how I map, build, and hand over human-reviewed automations.
Each of these is a live, operating automation — not a prototype. Architectures are described as built; metrics are pulled from real workflow runs.
A full social content engine for a craft cocktail tiki bar — built around a three-agent architecture where a Visual Producer orchestrates a Visual Director (image generation) and Visual Critic (quality scoring). The Critic returns an APPROVED / CONDITIONAL / REJECTED verdict and a regeneration note that loops back to the Director, with a hard cap at 2 regeneration attempts. Photorealistic venue imagery is generated by GPT Image 2, captions are routed by post type through dedicated agent prompts, and every post is gated through Telegram for human approval before Buffer schedules it. A custom Netlify dashboard replaces Retool as the client-facing UI.
AutoFlow — Twin City Automation (internal)
A complete n8n production pipeline targeting small business owners — runs from topic ideation through script generation (Gemini 2.5 Flash), voiceover (ElevenLabs), AI b-roll (Kie.ai), video assembly (Creatomate), and scheduled publishing through Buffer. Workflow ID ZjoONvK9bY6j7IgY is built end-to-end; final placeholders (Creatomate template, Buffer YT profile) are the only steps between the current state and full automation. Designed for 8 videos/month at ~$56 in API costs.
Wave — Charlotte & Atlanta
Cross-referenced 33 active drivers against FillHQ signed-document records and the 617-applicant hiring database. Surfaced that no driver was fully compliant — 10 had expired Insurance & Registration, 22 had no records on file, and Driver License expiry was never captured at onboarding. Delivered a management-ready Word report, a live-formula compliance tracker, an n8n CSV audit workflow, and a Zapier doc-renewal flow. Versioned through 9 iterations in production with full rollback capability — every change tested in a staging environment before promotion.
After delivering the compliance audit report to Wave management, the natural follow-up was: how do we prevent this from happening with the next hire? The answer was a production Zapier workflow with embedded AI document verification — currently running v9, used daily by Wave operations across both markets.
The master workflow: every new hire flows through three parallel verification pipelines — Registration, Insurance, Driver's License — each with AI-powered image validation, success/error branches, approve/decline gates, and downstream logging to Google Sheets, Airtable, Slack, and Gmail. Replaces a manual review process that used to take days.
The reusable verification sub-flow: a single document image is run through OpenAI vision to validate authenticity (registration / insurance / license), formatted with Zapier Formatter, then routed through Approve / Decline / Manual Review branches with full Airtable logging and email notifications back to applicants.
We Are Black America (WABA)
Two production pipelines: (1) a Supabase Edge Function-based news card compositor with a Creatomate video subworkflow and ElevenLabs/Qwen3-TTS voiceover; (2) a business directory enrichment pipeline using Tavily search and Perplexity/Qwen via OpenRouter. Plus a meme automation workflow against the Imgflip get_memes endpoint with fuzzy template matching.
Slappy's Chicken — Winston-Salem, NC
A social content engine with a dual-model image pipeline: Nano Banana for GPS-grounded outdoor restaurant shots, Qwen2 for blurred restaurant-feel backgrounds. Migrated text generation to Kimi K2.5 via OpenRouter for stronger food-voice copy. A Switch node routes Yelp-style customer review posts to dedicated agent prompts versus standard menu features.
Twin City Automation — infrastructure
A local Qwen3-TTS server running on WSL2 (RTX 3060, 12GB VRAM) and exposed to n8n through a permanent ngrok domain. Configured a custom voice profile (waba_hype) for branded voiceovers across WABA workflows — eliminates per-character API costs for high-volume TTS while keeping pipeline integration identical to ElevenLabs.
Twin City Automation — internal product
A structured catalog of every automation built across Wave, WABA, and other engagements — used as TCA's service menu and reusable building-block library. Each entry is tagged by domain (driver ops, content, recruiting, compliance), trigger type, and integration footprint, so any new client engagement can pull pre-built modules instead of starting cold.
Not workflows you'll abandon in 3 weeks. Systems your team can understand, operate, and grow with.
Every workflow is architected with the AI agent at the center — Producer/Director/Critic patterns, sub-agent tools, and memory passing — not a single GPT call duct-taped onto a Zap.
Anything that ships to a customer goes through a Telegram or email approval step. AI is fast — but my clients still review every post before it goes live.
Every node is named, every credential is documented, every cost is tracked. When I hand a system off, the next operator can run it without calling me.
n8n workflow exports backed up to git. Staging environment separate from production. Static sites auto-deploy from commit. Every system can roll back to a prior known-good version — because production incidents are inevitable and recovery time is what actually matters.
Currently taking on small business clients in Winston-Salem and remote engagements anywhere in the US.
I'll get back to you within 24 hours at the email you provided.