1 of 12

StructurAI: Intelligent Writing Support

Senior Design Project by Young Kwang Choi and Richmond Yevudza

2 of 12

3 of 12

Project Overview

StructurAI: An AI-powered writing assistant.

Streamlines editing by organizing content according to user-provided outline.

Utilizes logprobs for text reorganization and embeddings for text matching.

Benefits a wide range of users: writers, students, professionals, and more.

�

4 of 12

DEMO

5 of 12

Understanding Logprobs and Embeddings

Logprobs

the measure of how well a sentence fits within a given context.

Embeddings

the meaning of a sentence as compared to the desired outline.

6 of 12

Leveraging Logprobs for Sentence Placement Suggestions

Calculate how well a sentence fits in the outline sections.
Determine the most suitable placement based on logprob scores.

How Logprobs Work in StructurAI:

Efficient and accurate identification of suitable sentence placements.
Streamlined organization of written content based on user-provided outline sections.

Benefits of Using Logprobs:

7 of 12

Utilizing Embeddings for Text Matching and Organization

Calculate distances between text segment embeddings and outline section embeddings.
Match and organize content based on meaning.

How Embeddings Work in StructurAI:

Accurate and efficient text matching based on semantic meaning.
Enhanced content organization aligned with the user's desired outline.

Benefits of Using Embeddings:

8 of 12

Design Norms: Empowering Writers with AI Assistance

Faith: Providing Suggestions, Not Imposing Solutions

Linguistic: User-centric Approach

Cultural: Enhancing the Writing Process, Not Replacing It

Ethical: Ethical AI Integration

YK�Now, let's discuss the design norms that shaped our project. We believe that AI should empower writers, not take away their creative control. With this in mind, we developed StructurAI to provide suggestions rather than impose solutions. Our tool respects the writer's creativity, giving them the final say in their content.

Our user-centric approach focuses on the needs of writers, students, and professionals. We strive to enhance the writing experience while preserving the user's unique voice and style. This way, StructurAI complements their efforts and offers valuable insights and recommendations, encouraging users to further engage with their content and refine it based on our suggestions.

We also emphasize ethical AI integration. During the development and implementation of StructurAI, we were conscious of user privacy and maintaining a balance between automation and human input. By doing so, we ensure that our tool enhances the writing process without replacing the essential human element in the craft of writing.

9 of 12

Obstacles and Overcoming Them

Understanding Logprobs and Embeddings

Balancing User Experience and AI Capabilities

Adapting to New Techniques and Tools

Debugging and Troubleshooting

YK

•As with any ambitious project, we faced several challenges during the development of StructurAI. One of the main obstacles was the complexity of understanding and effectively implementing logprobs and embeddings. To overcome this, we invested time in researching and learning about these concepts, experimenting with pre-trained models, and continuously refining our implementation.

•Another challenge was ensuring the AI respects the user's ownership of their writing. We wanted to provide valuable suggestions without taking control of their work. To achieve this balance, we focused on designing our tool to offer meaningful guidance while maintaining the flexibility for users to make their own decisions.

•Lastly, we had to keep up with the rapidly evolving field of AI and incorporating new techniques into the project. In order to achieve this we maintained a growth mindset, iterated on the project based on new insights, and embraced continuous learning. Also, through debugging, troubleshooting, and refining our algorithms, we were able to overcome these issues and create an efficient, user-friendly writing assistant.

�

10 of 12

Limitations and Future Improvements

Dependence on pre-trained models
Context understanding

Limitations:

Domain-specific models
User feedback integration
Use a threshold to filter out placements with low relevance.
Making the UI more appealing to the user.

Future Improvements:

YK

While our StructurAI has shown promising results, there are some limitations faced. Firstly, our current approach relies heavily on pre-trained models, which may not always be ideal for all types of content or domain-specific writing. Second, the AI may struggle with understanding context-specific issues, affecting the overall quality of text organization.

To address these limitations, we have several future improvements in mind. One is to develop or fine-tune domain-specific models tailored for various writing styles and subjects. This will help improve text organization accuracy and better cater to different user needs. Another improvement is to integrate user feedback into the system, allowing users to provide input on the text organization and helping the AI to refine its performance over time. We plan to implement a threshold to filter out placements with low relevance, further enhancing the overall organization of the content. Lastly, revamping the UI a bit to make it very intuitive for users will be very good for the overall effectiveness of this tool.

11 of 12

Acknowledgements

Professor Arnold (Project advisor)

OpenAI and Spacy

Calvin University

12 of 12

Thank you. �Questions?