----------------------------- **While I generate an answer, here are some sources that might be helpful**: * [Connecting Your Custom LLM to Vapi: A Comprehensive Guide]() * [Introduction]() * [Update Assistant]() * [Prompting Guide]() -----------------------------

### Answer Improving data extraction for email addresses, especially when they are spoken, involves enhancing the accuracy of the speech-to-text (STT) transcriptions and refining the handling of custom spellings in prompts. Vapi does offer robust customization options for handling such scenarios, including the ability to define sophisticated parsing rules or use custom models that better handle specifics like email addresses. Here’s a general approach you can consider to enhance the accuracy of email address extraction: 1. Utilize Custom Transcription Settings: You can adjust your assistant's transcription settings to better capture specific formats like email addresses. This may involve enabling smart formatting or custom keyword configurations to ensure that the transcriber pays more attention to terms that sound like email addresses [(1)]() . 2. Prompt Refinement: Ensure that the prompts given to the system clearly instruct the model to pay close attention to how names and spellings are articulated by the users. This might involve setting up your prompts to include a verification step where the assistant repeats the email address back to the user for confirmation [(2)]() . 3. Post-Processing of Transcripts: Implement a post-processing step where the extracted email addresses are verified or corrected using a custom script. This could involve regular expressions or a lookup in a database of known contacts to correct common misspellings or errors introduced during the speech-to-text process [(3)]() . 4. Feedback Loop: Incorporate a feedback mechanism where users can correct the spelling if it is wrong. This not only improves the current interaction but also provides data that can be used to train a

nd refine the model further [(2)]() . These strategies should help in refining the extraction of email addresses from conversational data, particularly when dealing with custom spellings as indicated in your conversation. For practical implementation and customized solutions, exploring Vapi’s documentation on transcription settings and prompt customization might provide further insights and technical details relevant to your specific requirements [(1)]() , [(2)]() .

Got it. Tagging @User for additional help.

We are just using their service and using the exact results provided by them.