KnowMyself: Learn about yourself

I often hear people saying: "I don't what do with my major / "I study X but I don't like my job."

Even when I listen hard and ask tons of questions to understand their passions and goals, it's not enough. I'm still struggling to give them good advice.

That's why we created KnowMyself

KnowMyself takes the user's interests, hobbies, and context to offer tailored advice and suggestions. It assists individuals in discovering their paths by providing personalized guidance based on their unique attributes. Powered by PartyRock.

User cases:

1. Student with a multidisciplinary focus: Jane is passionate about biology and computer science. She's unsure how to align her interests into a career path. KnowMyself can help Jane by analyzing her skills and suggesting career options that merge her passions, like Genetics, biostadistics or AI applied to drug discovery; all fields merge biology and computer science.

2. Career shift: John works in marketing, but wishes to shift into political marketing. KnowMyselfAI can help highlighting relevant campaigns, political organizations, and advocacy groups that may benefit from John's expertise.

3. Uncertain Job Seeker: Lisa is open-minded about her career but has no direction. KnowMyself can suggest diverse industries based on her skills and offer exploration in areas like healthcare, education, or sustainable energy, each with unique opportunities.

Challanges, perfomance.

1. Both humans and LLMs have something in common: Both prefer information in digestable chunks

Most of our problems were UX related. Users tend to write short or very large responses. We decided to go for step-context process. First issue then context. Users were more satifisfed with the LLM's answer.

Around perfomance, certain models are more descriptive (like Llama or Titan) while other consice and structured (Jurasic). PartyRock's website the performs better with the Jurassic Family. Most models crashes the website some seconds while processing.

1. Getting the user's inmediate needs

Not all the users are looking for the same advice. KnowMyself is aware and looks to be personalized for each user

While testing the product, we noticed that high school students and college students have different needs (it's not obvious, stay with us). While some don't have experience, others have an idea and need the "how" based on their profile.

As previously said, KnowMyselfAI goes with step-context process. That also applies with the structure itself. The users who need a brainstorming or some ideas, only go for the first and second task. While the user who wants detailed and how-to would continue with the third task. At the end, if the users have more questions, they would go with chat.

The Best fit for each task

NOTE: Humans prefer chunks. We look for structure answers in models

• First task involved a LLMs context-aware with structured answers. The best option was the Jurassic family for both requirements. But Ultra is the best at organizing information on bullet points, making it more digestable. Worst model was Command: Designed for a more prompt-straight approach. This model is not so context aware

• Second task needed to take context by previous model and add more details. While Jurrasic might get more descriptive, command is the best tool here. Noted that this LLMs have temperature (0.4) and Top P (0.1), so for more creative (but also restrictive vocabulary) tasks Command performs better.

Note: Taking into account the Top P parameter, It ensures that command is created for straighforward tasks

• Third task involces a context aware and structure answers. Jurrasic Ultra was used here.

• Chat: Llama 2 Chat is designed for it, mainly as it is interactive and the most context aware.

A BIG THANKS FOR THE TEAM

Thank you a lot for this opportunity, hope you like the product. We'll be on the future updates on Partyrock

Built With

• partyrock