Generative AI Prompt Builder

Node-based Gen-AI prompt building platform; currently part of SAP Business AI and SAP Build

SAP

SAP

SAP

B2B

B2B

B2B

UXR

UXR

UXR

UXD

UXD

UXD

Challenge

Text-based tasks take too long time, but there is no perfect solution to cater each situation

Opportunity

Develop a Gen-AI Promp Builder that allows to orchestrate multiple ML (machine learning) models and use any Generative AI LLMs (large language models) for business application usage

Team

So Yeon Kim (Me, UXD)
Nassim Bahloul (UXD)
SAP ICN US Team (PM, Developers)

Timeline

4 Months

Tool

Figma

Microsoft Teams

MIcrosoft Words

Task

UX Design

UI Design

Prototype

I omitted confidential information on this page to comply with my non-disclosure agreement. All information in this case study is my own and does not necessarily reflect the views of SAP.

I omitted confidential information on this page to comply with my non-disclosure agreement. All information in this case study is my own and does not necessarily reflect the views of SAP.

My personal thoughts and anecdotes will be quoted on the right panel.

My personal thoughts and anecdotes will be quoted on the right panel.

Working with people from different timezones taught me about time-management a lot.

Working with people from different timezones taught me about time-management a lot.

Background

Creating a Gen-AI platform for everyone, not just developers

Business communications generate vast unstructured data—emails, texts, notes, and call transcripts. Manually reviewing and responding is time-consuming for any user.


The project started as a quick prototype for SAP employees to write emails easily by connecting different machine learning models, but it lacked organized screens and navigation.


Another UX designer and I joined to make the platform more intuitive. We focused on creating clear interactions relations and screen designs for the alpha release.

The biggest challenge for us was misleading information; how do we guide users to filter false-positives?

The biggest challenge for us was misleading information; how do we guide users to filter false-positives?

Product Structure

Integrating multiple models for efficient text analysis

The platform has two key functions: 1) integrating various LLM models and 2) using them for text-based business tasks like email summarization.

LLMs & ML models

LLMs & ML models

Chat GPT

Cohere

Claude

SAP LLM

Inhouse LLMs

Etc

SAP Gen-AI Package Configuration

SAP Gen-AI Package Configuration

Customize multiple LLMs

in a 'Package' format

SAP Gen-AI Prompt Builder

SAP Gen-AI Prompt Builder

Build prompts leveraging 'Packages'

Business tasks

Business tasks

Email analysis

Document analysis

Chat analysis

Response generation

Automatic CS

Etc

The biggest challenge for us was misleading information; how do we guide users to filter false-positives?

The biggest challenge for us was misleading information; how do we guide users to filter false-positives?

Solution

Configuration: Connecting different LLM and ML models

Connect multiple models in 'Package'

Users can customize a package of language models on the Configuration page

Node-based 'Package' visualization

To intuitively show what models are connected, I designed the platform in the node layout

Prompt Builder: Utilizing 'Package' to analyze texts

Changing parameter

Users can add, customize, and delete each parameter

Auto and customized prompt building

Users can apply the settings to auto-create a prompt formula while also adding their own prompt by typing

Security check

The user can turn on/off the security check, which will examine whether the prompt will cause any issues; the service will provide the user with a suggestion when hovering on the detected issue

Running prompt to get the outcome

Users can provide text and run the prompt to see the output preview

Prompt Lobby & Store: Sharing Packages with colleagues

Lobby: A personalized library to view my prompts and templates

Used tag components to indicate prompt properties like LLM type and prompt usage

Store: a place to browse and add public prompts

Located a browsing feature with advanced filtering options

The biggest challenge for us was misleading information; how do we guide users to filter false-positives?

The biggest challenge for us was misleading information; how do we guide users to filter false-positives?

Reflection

Learning Gen-AI to Design for Gen-AI

Understanding the back-end was essential for creating a logical design. Unlike simply using Gen-AI outputs (e.g., ChatGPT responses), Communication Intelligence enables users to build and customize ML models and prompts.

Negotiable and Dynamic Screen Components

Managing large amounts of information was a challenge. A negotiable layout allowed personalization based on user needs, while dynamic components optimized space by adapting to scroll input—such as headers that fold automatically.

Visualizing ML Models with Nodes

Inspired by my creative background in Touch Designer and Unity Shader, I introduced a node-based interface. This approach allows users to visually connect elements, review their workflow, and better understand the impact of each selection.

Experience

UX Visual Designer

SAP

2024.05 — Current

UX Design Co-Op

Motorola Solutions

2024.02 — 2024.05

Design Technologist Intern

SAP

2023.05 — 2024.02

...and more

Get in Touch

I’m happy to collaborate!

Please reach out via LinkedIn.

Education

New York University, Tisch School of the Arts

BFA Interactive Media Arts, Game Design

2018 — 2024

...and Have a great day!
So Yeon Kim @2025
So Yeon Kim @2025
Experience

UX Visual Designer

SAP

2024.05 — Current

UX Design Co-Op

Motorola Solutions

2024.02 — 2024.05

Design Technologist Intern

SAP

2023.05 — 2024.02

...and more

Get in Touch

I’m happy to collaborate!

Please reach out via LinkedIn.

Education

New York University, Tisch School of the Arts

BFA Interactive Media Arts, Game Design

2018 — 2024

...and Have a great day!
So Yeon Kim @2025