AI automation engineer working with the Claude API, the WhatsApp Cloud API, n8n and the Google Workspace stack. I ship agents that take a real conversation, create the calendar event, log the row and send the confirmation — end to end. One is answering real leads on WhatsApp right now. Open it and talk to it.
One agent runs in production on a real WhatsApp number. The others are web-chat versions of the same idea, wired to real Google Calendar and Google Sheets accounts. Send any of them a message and watch it handle the whole booking.
My own agency’s sales agent on the WhatsApp Cloud API, answering real leads from paid Meta ads. It qualifies, quotes in the lead’s own currency, books the call, understands voice notes and hands the chat to a human when it should. Under the hood: webhook signature checks, per-number rate limiting, retries when the model fails, a permanent conversation archive and 37 test files. It will try to sell to you — that’s the point.
Books appointments straight into Google Calendar, sells from a 10-product catalog (sends the product photo inside the chat) and logs every interaction to Google Sheets.
Books cuts with real double-booking protection — it checks live availability before confirming, then writes the event to the owner’s calendar.
Handles patient intake conversations and walks a new patient through picking a slot, with the clinic’s tone and service list built in.
A 34-scene walkthrough of a 10-bedroom rental villa, shot at 120 MP and built by a Python pipeline that denoises each panorama and emits lighter tiers so it still opens on a phone over mobile data. Bilingual EN/ES, reskinned to the owner’s palette, with per-scene analytics that tell them which rooms sell the booking.
The WhatsApp agent runs in production for my own agency. The web-chat agents are demo environments I built to show the engine, not client systems — I’d rather you judge working software than a case study I wrote about myself. The same engine gets reskinned per industry from one codebase, which is why a new vertical takes days, not weeks.
Self-hosted, running every day, no public link — happy to walk you through any of them in writing.
Web form, WhatsApp and Instagram DMs land in one flow. Claude scores the lead, it’s written to a Google Sheets CRM with dedup by ID, hot leads trigger an alert and the rest get an automatic email follow-up.
Reels and stories go out to Instagram and Facebook from a calendar, through the Meta Graph API, with a retry branch that republishes whatever failed.
One call returns processed Instagram and Facebook numbers — by time slot, by post — so the weekly review doesn’t need anyone to log in anywhere.
Conversational agents that qualify, answer and book. Claude as the reasoning layer, the WhatsApp Cloud API as the channel, n8n for the plumbing around it.
Webhooks, API integrations, error handling and retries. Also: fixing and diagnosing workflows someone else left broken.
Calendar, Sheets and Gmail wired into the flow — service accounts, OAuth2, idempotent writes.
Document and image parsing, structured extraction, classification. I pick parser-vs-LLM per field instead of sending everything to a model.
Video generated from React code. Full pipeline: voiceover → Whisper transcription → SRT timestamps → scene cuts → rendered MP4. My agency’s reels come out of it.
React 19, Tailwind v4, Framer Motion. Fast, responsive, deployed on Vercel.
A message, a Loom, a broken workflow. No discovery call needed to get started.
Fixed price for defined work, hourly for open-ended. You approve before I touch anything.
I deploy to a live URL you can open and test yourself — same as the demos above.
Workflow JSON, credentials setup and a short written handover. The system is yours, not rented from me.
Tell me what the manual process looks like today and I'll tell you honestly whether it's worth automating — and roughly what it would cost. Written English works best for me; I reply within a few hours.