I build automations that save time and cut costs.

A home service business's phone is its front door, but nobody's free to answer it around the clock. Technicians can't stop mid-repair for a second call, and nobody's picking up at 9pm when a pipe bursts. Every call that goes to voicemail is a real chance the customer just calls the next name on the list. And when someone does answer, checking the calendar, confirming a slot is actually open, and getting the customer into the CRM all happens by hand — which means it depends on whoever picked up writing it down correctly and not double-booking a tech who's already out on a job.

Vapi answers the call and handles the conversation, while n8n runs everything behind it. When a caller asks for an appointment, the system checks Google Calendar in real time, works out open slots around business hours and job length, and offers the caller real options. Once they pick a time, it re-checks that the exact slot is still open before booking anything, creates the calendar event, and only then syncs the customer and job into HubSpot. If that CRM sync fails after the calendar is already booked, the appointment stays booked — the failure gets logged and a human gets alerted, instead of the system rolling back and losing the job. Callers can ask for a real person at any point, which escalates with the callback number and job details already attached. Every call is logged with its outcome, and a dead-letter system catches failures anywhere in the pipeline so nothing fails silently.

Once an assessment wraps and a client's ready to move forward, onboarding still means someone building the ClickUp task by hand, drafting the service agreement, and following up until it's signed. There's no check for a duplicate request coming through twice, and whether the paperwork actually went out lives in someone's memory instead of the system.

The moment the care team marks a client ready, n8n verifies the request, checks the record for what onboarding needs, and stops anything that's already been processed. It creates the ClickUp task and sends a DocuSign agreement pre-filled with the client's details. When the signature comes back, a second flow confirms it, pulls the matching record, saves the signed file to the client's Drive folder, and updates the ClickUp task and client record to match.

Manual dispatch turns a simple booking into a chain of phone calls and guesswork. Someone has to check who's available, decide who's the best fit, offer them the job, and start over if nobody answers. It's easy to lose track of who's already been asked, or where a job actually stands.

Built an automated dispatch and job-tracking system in GoHighLevel and n8n. When a job is marked Booked, n8n checks technician availability, workload, and capacity, then offers the job through Telegram with a set response window. A decline or no response escalates to the next eligible technician without duplicate offers. On completion, the technician sends a voice note. It gets transcribed and checked by AI for completeness before the job closes out in the CRM.

Support answers live scattered across documents, guides, and internal files. Finding the right one takes longer than it should, and a standard chatbot without access to that material just guesses.

Built a RAG support system in n8n that answers from the company's actual documents. Source files are chunked and embedded into Pinecone. On a question, the system retrieves the closest matches and reranks them with Cohere, since raw similarity search alone tends to surface mediocre results, then generates an answer from that context. Conversation memory runs through Postgres, so the chatbot remembers what was already asked earlier in the same conversation.

When a Magento development contract closes, the account manager loses billable hours to setup work. Building folder structures by hand, drafting the contract, looping in engineering: all of it delays the actual kickoff.

Built an automation that fires the moment a deal is marked Closed Won in HubSpot. It creates the Drive folder structure (Contracts, Project Files, Invoices), generates the contract doc from the deal's own data, and shares the folder with the client, all at once instead of one step after another. In parallel, it sends the welcome email, logs the client in the tracking sheet, and notifies the dev team in Slack.

Technical support inboxes mix everything together. A critical alert (a proxy failure, a blocked automation, a backend error) sits in the same inbox as routine account questions and spam. Nothing marks it as urgent, so it waits its turn.

Built a triage system that reads every incoming Gmail message and classifies it by urgency before a human sees it. Spam gets ignored. Routine questions get an AI-drafted reply, held in Gmail for a human to check before sending. Anything urgent skips the queue and pings the engineering team's Slack channel right away.