Design the AI use-cases

[hungdtrn]

Before designing the prompt or AI functions, I think we should be clear of what we are building. What can this AI bot? And how users will use this AI?

[hungdtrn]

I think the general chat bot is not much helpful. I could not think of a use-case where users might use the QA bot in discord.

The potential useful cases are:

• The user ask the bot about information in the Discord server.
• The user ask the bot to revise the CV, correct grammar mistakes and give general comments
• The user have a conversation with the bot to revise a CV

Examples:

1. The user ask the bot about information in the Discord server.

   • I want to know about the expected salary of ML engineer. I know that we have a thread for users to share about salary. But I want a faster way. I ask the bot to help me.
   • I want to know who is working as a ML engineer in the discord server. I can ask the bot to help me as well.

2. The user ask the bot to revise the CV, correct grammar mistakes and give general comments When I ask for reviewing CV, a bot will be triggered to skim through my cv, correct grammars and give general comments. Then other users can provide more detailed feed backs.

3. User have a conversation with the bot to revise the CV

   • I uploads the CV, then the bot give comments about sections in the CV.
   • I then ask how to revise, the bot provides some ideas.
   • The conversation can continue until I am satisfied with the new CV

[dacphuc1993]

Since the first use case requires storing server messages, thus causing privacy concern, we can rule it out for now.

The second and third use cases are closely related. Given the scope of this project, I think we should set the second one (reviewing CV) as a deliverable and put the third one as optional.

If any use case arise during the project, we can add it along the way should members see it fit.

[hungdtrn]

Yeah, that make sense. Thanks a @dacphuc1993 In the discord server, the product owner still want to keep the general OpenAI QA use-case. Since the general QA use-case is simple, I think it's fine to keep it.

[AndrewsTrinh]

Thanks @hungdtrn and @dacphuc1993. I have added 3 new use case to the document, including:

1. An AI that act as Interviewer in a role play interview with user
   • When user prompting the command:
   • The bot create new thread inside channel and naming it like interview-role-play-
   • The bot act as interviewer and do roleplaying with the user
   • Other user can assign their role as co-interviewer and join the interview
2. An AI that generate initial troubleshooting idea when user post their issue
   • A kinda sub use case of use case 1
   • User ask abot their issue on the channel. Prompt to generate suggested solution
3. An AI that summary long post
   • User reply to the long post with command to shorten to summary Can you please review these usecase if they are feasible and make sense?

[dacphuc1993]

1. Simulating interview I don't think interviewing by chatbot verbally (texting) will be an attracting feature unless it's fully automated by voice (which mean user speak to microphone and the system generate audio responses). It is quite difficult for a project of this size if we have to add another text-to-speech module. We can leave this for later stages.

2. Troubleshooter I am not sure I understand the use case. Currently, we have /ask command to ask general questions. If an user need to find a solution to a problem, they can use /ask and describe it with a prompt. Maybe I miss the point of this feature?

3. Summarize text I think this is a great idea. The users that might be interested in this feature are the ones without ChatGPT-plus. So if we support uploading files (pdf, docx...) and summarize a file, it can be convenient. OpenAI API supports parsing files afaik. We can discuss further with other members to see if it is worth adding to this phase as we might have to add another slash command /summarize and handle different data types/length.

[nqngo]

Hi folks, I greatly appreciate the effort planning here and my apology for not replying earlier in the conversation. Our use case for this project at the moment are pretty clear and limited:

1. /ask is pretty much an interface to OpenAI/Ollama. We pass the whole prompt to the endpoint and whatever returns we give it back to the user. This is great because we will not be focusing too much on the "prompt engineering" part of the implementation. Our goal is very much present the capabilities for testing or quick general knowledge answer to mimic the user experience when using ChatGPT. This is why it is our first issue.
2. The second component is the resume review part of VAIT. I want to be able to provide a quick feedback cycle for members who trust us to provide constructive criticisms on their resume. Please refer to previous resume review such as: https://discord.com/channels/784198833815027742/1207868289569783908 https://discord.com/channels/784198833815027742/1211265254819110942 https://discord.com/channels/784198833815027742/1206097495793475614 https://discord.com/channels/784198833815027742/1162753669087907892

Our goal for number 2 use case is not replacing the human experience, but get the bot to provide general feedbacks and catches any easy errors.

For other use cases brainstormed in this issue. I think it will be best to address after the main project wrap up, if everyone else is keen on continuing the maintenance and development of our capabilities.

[samhwang]

Are we done with this or it should be converted into a Discussion?