I Build AI Systems That Automate Business Operations

Answers lived in eight separate systems, so every routine question turned into a manual report someone had to build.

An internal assistant that queries the medical record, billing, CRM, ads, calls and content data on demand, briefs each role every morning, and dispatches work to the specialist agents below.

Prescriptions ran out and labs went overdue between visits because nobody was tracking the gaps.

A timeline that syncs from the medical record, calculates each run-out date (pharmacy delay included), and flags what needs attention two weeks early across four role-based dashboards.

Providers wrote up consult notes after hours, from memory.

Browser-based consult capture → transcription → an AI-drafted clinical note, care plan and patient handout that the provider reviews, edits and signs.

Producing on-brand social content — copy, visuals, and video — was a daily manual grind across multiple tools and people.

A 4-stage n8n pipeline: AI drafts the post → auto-generates the visual → routes to a human for approval → auto-publishes on schedule.

The team guessed at what to post and reacted to trends days too late.

An agent that scrapes competitors and trends, scores ideas against the brand’s content pillars (engage / counter-program / ignore), and feeds the content engine.

Editing provider videos — captions, music, voice-clone — tied up a laptop for hours per clip.

A containerized Python/FFmpeg editor on Google Cloud Run that auto-captions, scores, and renders short-form video on demand.

Paid acquisition meant a media buyer hand-producing concepts, checking each one for compliance, then building campaigns manually.

An agent that studies competitor ads, drafts pre-screened ad concepts, and — only after a human approves — creates and launches the campaigns behind hard budget caps.

Inbound calls needed answering 24/7, and robocalls were burning per-minute cost and reaching the founder’s phone.

A voice agent that answers, qualifies and books straight into the CRM calendar — plus a pre-answer screen that rejects known robocall numbers before the call ever connects.

Provider intake ran through a clumsy Google Form, and providers had no visibility into their own panel or pay.

A branded portal that replaced the form, plus a per-provider dashboard showing panel size, churn and monthly earnings — synced nightly from the practice’s own records.

Matching members to the right learning pod was manual, slow, and inconsistent.

A GoHighLevel Conversation-AI bot that interviews members and returns ranked pod matches as cleanly formatted match cards.

Members didn’t know where to go for help and questions piled up.

An always-on assistant bot that answers questions and routes members to the right resource or a human when needed.

They needed a branded member area without page-builder bloat or lock-in.

A hand-coded BuddyBoss child theme — member profiles, course and community UI — deployed safely via WPCode.

Member profile data lived in WordPress; sales and SMS lived in GoHighLevel.

Synced WooCommerce Memberships and BuddyBoss profile fields into GoHighLevel and Twilio for automated, personalized outreach.

A sprawl of expensive, multi-step Zaps that kept hitting task limits and were hard to maintain.

Rebuilt the critical Zaps as clean, version-controlled n8n workflows — lower cost, more logic, far easier to debug.

Invoices were read by hand and re-keyed into accounting — slow and error-prone.

Email-in invoices are parsed by AI, deduped by invoice number + vendor, and pushed to Notion and accounting automatically.

Insights from sales calls never reliably made it into the CRM.

Fireflies.ai transcripts flow through n8n into Close CRM with AI summaries and auto-created follow-up tasks.

Tracking competitors’ Instagram content was a manual, every-week chore.

Apify scrapes posts and reels, AssemblyAI transcribes the audio, and the results land structured in Google Sheets for review.

AI voice agents couldn’t reliably book, reschedule, or cancel inside clients’ GoHighLevel accounts.

A custom MCP server that exposes GHL calendar and contact actions to voice agents — multi-tenant across every sub-account.

Agent configurations silently drifted and broke in production with no warning.

A healer that detects drift and auto-restores each agent to a known-good state, with a full event log for auditing.

After-hours calls meant missed bookings and lost revenue.

A voice agent that answers, qualifies the caller, and books the appointment straight into the GoHighLevel calendar.

The agency needed a way to manage agents on behalf of each client.

An admin API powering a client-facing portal for provisioning and managing voice agents per account.

Nobody could see how the deployed voice agents were actually performing across every client account.

A fleet dashboard over all call data — volume trends, after-hours share, sentiment, and a “needs attention” feed — where every number drills through to the calls behind it.

Reception was overwhelmed and after-hours calls went unanswered.

An AI receptionist that handles routine enquiries and cleanly hands off to staff the moment a human is needed.

The booking source of truth is the clinic’s PMS (Cliniko), not the CRM — and they must never disagree.

A Cloudflare Worker relay that mirrors PMS data into GoHighLevel while keeping the clinic’s system as master.

Posts either sounded generic or took hours to write. Anything AI-drafted read as AI, so it never got published.

An agent trained on the founder’s ratified brand voice, banned-word list and beliefs that writes essays, carousels and reels every week — in that voice, on those topics.

With no honest source of personal anecdotes, the agent invented them — the fastest way to lose a founder’s trust.

A private journal for raw, first-hand reflections; the agent extracts the themes and may only draw personal stories from that bank. If it isn’t in the bank, it doesn’t get written.

Turning an approved script into a finished vertical video meant an editor and a week of back-and-forth.

Three render modes from an approved draft in about two minutes — captions over b-roll, a cloned voice over imagery, or the founder on camera.

Every piece had to be personally approved — without the founder living inside another dashboard.

A chat-first portal with a review queue: approve, ask for a rewrite in a different tone, or decline with undo. Nothing publishes without a sign-off.

Renewals were chased by hand across eight membership types, each with its own pricing, cadence and paperwork.

One CRM workflow that branches by membership type and runs a timed reminder cadence — 45, 30, 15 and 7 days out — with the right template and that member’s own invoice link in every email.

Two member categories are priced per employee or per record filed, so their invoices had to be calculated manually every year.

A webhook flow that reads each member’s live counts at send time, calculates the amount, generates the invoice and writes the payment link straight back onto the contact record.

New member applications were approved over scattered email threads with no clean record of the decision.

An automated vote flow that collects board decisions, tallies them, and moves the applicant forward — or back — automatically.

Inconsistent member data meant renewal sequences and segments kept targeting the wrong people.

Rebuilt the member data model — types, statuses, renewal dates — so every downstream sequence targets accurately instead of guessing.

Every client offer was rebuilt from scratch, so launches were slow and no two accounts were run the same way.

A documented launch process turned into a reusable project template — offer, funnel, calendar, copy, workflows, testing, post-launch — so anyone on the team can run a launch to the same standard.

A failed launch traced back to a booking-calendar mismatch nobody had checked — the kind of mistake that repeats until it’s written down.

A ten-stage SMS campaign template with every hard-won check built in, from calendar setup to list filtering to a live send test. Reusable across every sub-account.

Campaigns were texting people who had already booked the offer.

Segmentation that strips out booked and converted contacts before send, so the list only reaches people the offer still applies to.

Owners couldn’t see what the marketing actually produced month to month.

Multi-offer funnels and booking flows per account, plus a standard monthly report pulled from the CRM — leads, bookings and revenue attributed per offer.