Latin in Year 5 is... well, lots of text, little context. My son found it rather dry. So I wanted to see if I could change that with a little AI-powered workflow automation. The idea: He should be able to take a photo of his exercise page with his iPad – and it would automatically turn into a colourful comic strip containing exactly the same vocabulary and grammar points. So I built a little web app: Select photo → upload → press ‘Crea!’. An n8n workflow then runs in the background → result sent back to the iPad via push notification. A little later, a message arrives: ‘Your Latin illustration is ready!’ I even deliberately included a little hint: ‘We've hidden some mistakes... can you find them?’ Because the AI sometimes draws details differently, the bug becomes a little learning feature. He searches, compares, reads – and repeats the vocabulary along the way. The whole thing runs via a compact n8n workflow: Image from the web app → cache → prompt extension → image generation → Pushover to the iPad. The visualisation comes from Nano Banana Pro. I wanted to test how well the newer models interpret complex scenes from exercise pages – it works surprisingly well for didactic mini-cases.

This will help everyone know which is most used to find which is best.

Moin aus dem schönen Ostfriesland! ich würde mich über Kontakt zu deutschsprachigen Teilnehmer:innen freuen. Mich interessiert u. A., warum Ihr dabei seid, welche Ziele Ihr verfolgt, und welches Potenzial Ihr für KI im deutschsprachigen Raum seht. Lat us mal proten ☺️ Servus Roland

Building my first RAG I'm finally able to complete the very first component of my first RAG. So far, I'm beginning to understand that the success of the workflow's execution is in the correct configurations of the moving parts. My challenges around the configuration is on making the Google Drive - Download File node work. I hit many walls trying to make it work with OAuth2, but for some reason, my access gets blocked. So because I want to succeed by this time, I took a pivot. I tried connecting via Service Account and that simplified it. With this though, manually adding the service account (Google Cloud) to the file as an Editor is necessary, so the expected file that you desire to retrieve is returned in the n8n dialog. With Pinecone, I hit another wall, my execution returned a dimension mismatch error between the index and the vector, so I had to recreate my index. Today was indeed a mental workout 😅 Up next is the second component of this RAG - which is the agent, will share more about that next time. I'm planning to connect this with a personal editor/content refiner agent for my personal blogs, that will return my dumped ideas into ready content that I can just review and then post 🤩 Let me know in the comments what are your challenges, interested to learn from you as well 😍 #happyBuilding