I automate the back office I used to run by hand.

WhatsApp assistant for a gym

Espacio Solé, Gonnet, Argentina. 2,400 members.

In production since June 2026

Members ask about schedules, fees and classes on WhatsApp. The front desk was answering all of it by hand. The assistant answers on its own and only hands over to staff when a sale is ready to close.

• Started as a 51-node n8n workflow; migrated to native tool use on the Claude API to make it simpler and more reliable.
• Knowledge base with semantic retrieval and prompt caching; answers stay inside what the gym actually offers.
• Postgres on Supabase with separate schemas for the bot runtime and the admin panel, row-level security on every table.
• Every change is tested against a regression suite of 295 audited real conversations before it ships.

Claude API, n8n (self-hosted, Docker), WhatsApp Cloud API, Supabase, Next.js admin panel

Finance automation for a student residence

Casa Rica Obelisco, Buenos Aires. I was the administrator.

In daily use

I handled the company’s treasury: payments, transfers, collections and the month-end close. Instead of living in spreadsheets, I automated the parts that repeated every month.

• Automatic ingestion of bank movements into the accounting system, including parsing credit-card statements from PDF.
• Invoice files renamed and filed automatically by an n8n workflow.
• Replaced the cash-flow spreadsheet with a web app that projects the bank balance day by day, so the team knows ahead of time how much money is needed and when.

n8n, Python, Next.js, Supabase, Vercel

Visit booking and referrals for a coliving

Wanna Colive, Argentina.

In production

Prospects used to book visits by message, and follow-up depended on someone remembering. The platform schedules visits, tracks each prospect through to move-in, and notifies the sales team automatically. A referral system rewards residents who bring new ones.

Next.js, Supabase with RLS, n8n orchestration, Vercel