Automatizaciones
In Developmentn8n workflows I built to practice process automation: a daily exchange rate and an AI message classifier
Role: Author
Solo · Since Oct 2026 · Public repo
Two n8n workflows with their own repo. One pulls the USD to DOP rate every weekday, stores it in Postgres and compares it with the previous day. The other is a webhook that sends a customer message to Gemini and returns its category, urgency and a one-line summary.
Problem
A language model can answer with the wrong format or with whatever the customer's message tells it to, and a workflow that trusts that output breaks quietly.
Solution
The customer's text goes between tags and is treated as data, the answer is checked against the allowed values, and the webhook returns 400 for bad input and 502 when Gemini fails after three tries.
Architecture highlights
- 6 tests against the real webhook, including a message that tries to give orders to the model and a forced 502
- The exchange rate lives in a Postgres table with one row per day, so manual runs and scheduled runs see the same history
- A GitHub Actions job checks that every exported flow imports cleanly and carries no keys
- Postgres in Docker instead of my SQL Server Express, which only accepts Windows authentication: the reasoning is in the repo's decisions log