Testing Lindy: Challenges with Email Triager and Model Performance
Greetings, we are testing Lindy to see if it can produce results. Currently even the simple Email Triager is not working well. For example, the inbound email clearly is addressed to myself ("Hi Myles") yet the Condition Met for labeling the email response was: "Condition met: it this email appears to be relevant to a company project or community initiative but doesn't directly mentioned Myles or no action appears to be required." leading to incorrect label. I was using ChatGPT 4.1 Nano so I have changed it to Google Gemini Flash 2.0 to see if it performs better. Any advice on what models are working for y'all for each step of the Triage workflow?