I'm bashing my head against the wall trying to solve this problem with LLMs. I've tried over 10 LLMs now with a varity of different prompts, but all give the same response. Prompt: Analyze the following transcript and extract the email address, taking into account any corrections or clarifications made during the conversation. Pay attention to the speaker's name and any potential typos or mishearings. You will likely see what the spelling of an email address is, but you should look at the rest of the conversation first to determine what the correct spelling is. Return the extracted email address in JSON format, with the variable name "user.email". Transcript: hello how can i help? i want to give my email and i spell my name with a K and not a C Great what is it? chrismanco at hotmail dot com Result: { "user.email": "chriskmanco@hotmail.com" } Other result: Result: { "user.email": "chrismanco@hotmail.com" } Expected Result: { "user.email": "chrismanko@hotmail.com" } Other potential expected result: { "user.email": "Khrismanco@hotmail.com" } Anybody got a better way to prompt to get the desired result?