02Work
UGC-style spots, TikTok-native edits and AI-assisted video for Meta and TikTok, where you get about a second before the thumb moves on. I make these in-house because I also run the ad accounts, so what the numbers say on Monday shows up in the next edit.
I write several hooks for one product, cut each into a UGC-style or AI-assisted edit, and launch them side by side in one campaign. Losing hooks get switched off fast and cheap. The winner gets budget the same week, and whatever it taught us goes into the next batch. The reels below show the formats I cut: UGC-style, promo, GRWM, product demo, vlog and fashion edits. Swipe for more.
Posts and product graphics that keep a brand's feed selling between videos. Ten skincare pieces here: benefit callouts, a serum launch, a buy-one-get-one and a 50% off pre-order. Swipe for more.
Nine systems that run the busywork behind the ads, the store and a creator program. The first three open to their n8n canvas; the rest are the systems I built for The Loaded Teashop at AMA Holdings, drawn step by step.
Answers customer messages on three brand pages. Claude reads the thread and the brand knowledge, a compliance filter checks every reply, and the moment a human answers, the bot pauses on that thread and the team gets a Telegram alert.
Watches a Drive folder for supplier packing lists, PDF or CSV. It maps supplier SKUs to the master list, validates every line, blocks a list that was already received, then emails a human to approve before it posts the stock-in and logs the batch. Anything it can't read goes to a review folder with the reason.
Every hour it reads the replies in Gmail, pulls the lead list from a sheet and classifies each lead. New leads get a three-email sequence written by one GPT-4o model and checked by a second; warm ones get the follow-up; anyone who replied is marked so no one gets emailed twice.
For The Loaded Teashop at AMA Holdings. New creator uploads reach a QA queue within a minute, with flavor, funnel stage and whitelist handles filled in. Every 15 minutes the approved ones are filed in Drive by day, flavor, creator and funnel stage, and written to the media buyers' sheet with the right ad copy. Failures write their own row.
Every 15 minutes: spend, blended ROAS, Triple Whale and Northbeam ROAS and CPA for every video, judged against the team's targets, into the dashboard's Ads tab with time frames from today to all time. A failed run turns its card red instead of going quiet.
Hourly. The creator platform's Meta assets fill in missing Instagram and Facebook handles for 833 creators and mark who is set up. The pipeline reads the same table, so a creator with no whitelist holds their videos until it is fixed. Team edits are never overwritten.
The team's second brain as a chat: call transcripts, concepts, audits, playbooks and the program's ad numbers. Claude searches, reads and cites the notes, flags advice older than six weeks, saves what teammates tell it back into the vault and can hand over data as a new Google Sheet.
A chat command. Apify searches TikTok and Instagram, Claude scores each creator against the ideal creator profile, the good fits land in a sheet and the shortlist comes back in the chat that asked.
Raw phone clips in, a finished vertical ad out. Deepgram writes the transcript, Claude plans the cut and the text cards and labels the shots, ffmpeg renders. Roles and monthly limits per member; it runs on a Mac mini behind a tunnel with a Vercel front door.