Querying Your Emails

Now that you’re emails are loaded into the tool, let’s get to harnessing the true power of the pen!

Below we have not only the primary component of our QuillQuery tool, but also step-by-step instructions for those that are not familiar yet with the functionality or some of the features.

Let’s Get Querying ...

Step 3: Get to Querying

Now comes the fun part, querying your loaded emails! There are a few inputs, however, you’ll need to submit for this step.

Input 1: Your Query

We deliberately use the word querying rather than searching as the user experience of QuillQuery is slightly different than your traditional email searches. Rather than typing in keywords, for our tool requires a more conversational input. For example, rather than searching for the keyword ‘commencement’ to get the date of your graduation; you can submit the query ‘when is graduation’ or ‘what is the date of my graduation’. Perhaps you want to better understand more broadly about all the elements of graduation; you can submit the query ‘give me all the details regarding my graduation to include the date’.

Input 2: Your Choice of Language Model

As we’ve mentioned, our tool we give YOU the option of what type of language model you want to use based upon your specific needs for your specific query or preference.

Perhaps you are a magical Microsoft user and really like the Copilot and ChatGPT experience, or maybe you love Gemini from Gmail. You may even be a connoisseur of Claude, which is not available for any of the current email applications. Also, choosing which model to use is somewhat a matter of personal choice, but can also be impacted by the types of text in your emails. For example, are your emails typically shorthand and phrases, or are they more formal with complete sentences.

Well have no fear, you can chose to whichever model you like best, and we’ll take care of the rest!

Currently, we provide the following model options: Cohere R, ChatGPT GPT-4o, Anthropic Claude 3.5, Mistral 7B Instruct, Meta Llama 3.1 8B Instruct.

If those names mean nothing to you, have no fear! We’ve provided a bit breakdown on how each model is different, what are good use cases for that model, and an example output so you can see the types of responses to the query “When is Fall graduation for MIDS?”.

• ChatGPT GPT-4o - Most conversational. Provides added context to response. Most capable of complex, nuanced language. Best at analyzing shorthand. “Graduation UC Berkeley Fall 2024 MIDS graduating class is on 21 December 2024 from 5pm to 6pm at the Haas Pavilion. Be aware that the deadline to register is 20 November 2024 and there are no extensions.”
• Claude Sonnet 3.5 - Most conversational. Typically identifies the specific source within the email. Most capable of complex, nuanced language. Best at analyzing shorthand. “There are two emails referencing graduation: one sent on 10 Nov and a second on 18 Nov. According to both, graduation is 21 December 2024 from 5pm to 6pm.”
• Cohere R - More conversational, but not as smooth as GPT or Claude. More direct responses with less context. More capable of complex, nuanced language than most models. Better at analyzing shorthand. “Graduation for the UC Berkeley Fall 2024 MIDS graduating class is on 21 December 2024 at 5pm.”
• Mistral 7B Instruct - Most direct responses with no additional context or information. More susceptible to errors with nuances or conflicting information. Formatting not as consistent as GPT, Claude, and Cohere. “Answer: Graduation is on 21 December 2024.”
• Meta Llama 3.1 8B Instruct - Most direct responses with no additional context or information. More susceptible to errors with nuances or conflicting information. Often produces formatting errors. “graduation is on 21 December 2024.”

Input 3: Submit Your Query

Your final input for this section is to hit the submit button!

Most users are interested in getting back a list of emails that they can review, but more and more show interest in extracts or even summarizations of emails to help them synthesize information faster. As such, we give you both the summarization of the contents as well as the list of emails for you to review.

Step 4: Log Out of QuillQuery

To conclude your journey in QuillQuery, you’ll need to logout of the tool. Logging out deletes all of your personal data and custom-built architecture. This ensures maximum security of your data and privacy.

Of note, in future releases, we plan to incorporate a pay-for-use option in which you will be able to maintain your data and user-specific pipeline to remove the need to upload your data each time you use the tool.