Park Joo Young

Generative AI tells lies.
Why? Because they collect data from everywhere.
So, this project began with the question: What if only "relable" information is input into the AI's database?

We defined "relable information" as "academic papers” published in reputable journals and started the project by interviewing academic researchers.

We categoried interviewees and created personas into four types as follows:

1. Undergraduate students
2. Ph.D. students
3. Professors at universities with limited access to research assistants
4. Professors at prestigious universities

We defined their specific "profiles," "goals," and "challenges" based on user interviews and secondary research, identifying the "pain points" in the current market.

Competitor analysis was conducted on "generative AI" market,
comparing indirect competitors, which are general knowledge-based conversational services targeting the general public, and direct competitors targeting academic researchers.

The analysis revealed that general services were vulnerable in key features such as "reliability," "security," and "up-to-dateness" of the "questions and answers."
Additionally services targeting academic researchers could not address the functional difficulties of our personas.

We designed the user flow by referencing the flow of general services, using a search bar to make users input the prompt easily.

Brand identity and content tone were designed to be friendly and memorable, referencing the strategy of the smaller company "C."

The pain points identified through persona definition and competitor analysis were as follows:

1. Conversational AI services were not suitable for assisting academic researchers due to the mixture of "non-expert knowledge" esearch Assistant”.
2. Conversational AI services were also vulnerable to security because they learn from user questions.
3. the update cycle of data in conversational AI services was fixed, weakening the "up-to-dateness" of responses.
4. Current AI-Assistants for academic researchers could not provide as detailed and analytical answers as conversational AI services.

We listed solutions to address each pain point and defined the value proposition of our service.
The four values became the unique value of "Tlooto" that differentiates it from other services.

However, the feature to write "detailed and analytical report" needed to be technically addressed, so we decided to monitor the development team's progress.

The test results were as follows:

1. Most people did not immediately understand the service's main feature from the [Homepage].
2. However, they could easily input prompts.
3. The drop-off rate at the login stage was high, making it difficult to reach the [Report page].
4. Users found it challenging to create prompts according to their academic needs.
5. While most users were satisfied with the quality of the reports (Technical constraints was addressed), many suggested that the importance of the papers be reflected when AI selected the list of citation.
6. There was negative feedback on the readability of report in the [Report page].
7. Many people clicked the [Full text] button on the [Report page].
8. Many people also clicked the [Full text] button on the [Paper page].

Based on the test results, we revised as follows:.

1. We separated the [Homepage] and [Search page], and redesigned the [Homepage] to highlight its main features.
2. The [Report page] can be shown before logging in or signing up in order to motivate users to log in or sign up.
3. Through user interviews, we found that users specifically wanted AI to assist them in finding research topic, writing contents of tables and litertature review, and deciding reseach methodology
4. We added prompt examples of each of four specific purpose in [Search page] and enable users to see the results (reports) of each example prompts before loggin in." 5. We revised the layout of the [Report page].
6. To save time, we removed the [Paper page].

We revised the user flow to reflect user feedback.

In particular, we ensured that users could see examples and results of prompts related to research topics, contents of tables, literature reviews, and research methodologies even before logging in.

We designed the provided prompts and report as "fixed contents", ensuring that the server does not have to work heavily, thereby minimizing maintenance costs.

Based on user research, we have added a "paper ranking algorithm."