Nov 2024 - Dec 2024
Teammate
Tomas Alberto
Mentor
Adam Bruke
UIUX Designer, Front-end Developer
A Figma Plugin and Web-based product
Project Objective
An AI tool that can intake text data or transcribed dialogue from a design review and convert it into a detailed overview of keywords, summary, and sentiment graph.
OUR VISIONImproving the personal experience of the daily working routine and process as a UX/UI Designer using natural language processing (NLP).
OUR MISSIONTo automate and quantify the often subjective practice of feedback and review for the UX/UI Designer, which can increase efficiency in making improvements.
Our solution, SentimentAI, is an AI tool designed to analyze text data or transcribed raw dialogue from client-approved design reviews. It generates a comprehensive summary, including keyword extraction, sentiment analysis, and design categorization, allowing designers to pinpoint specific elements for refinement.
Our target audiences are UX/UI and product designers, particularly those that interface with various clients such as those at an agency or through freelance. The tool can be hosted on a live web server so that users can access it from different devices.
In designing the user flow for SentimentAI, I aimed to create a clear and intuitive process that mirrors how designers naturally work with client feedback. The flow starts with collecting original text or voice input from client-approved reviews, which is then labeled and stored for reference. Within the audio or text input, the NLP will process the original text and extract the keyword. The keyword will help designers to learn the insights from clients that related to the design vocabulary. The model will then summarize such insights into a unified meeting report for a readible version in order to review afterwards. Meanwhile, the sentiment scoring occur in parallel to provide both detailed critiques and emotional context, helping designers to precisely capture the attitude of clients.
NLP Model
The product used three open-sourced natural language processing models from hugging face.
APP Workflow
Low Fidelity Prototyping
High Fidelity & Deployment Version One - Figma Prototype
Input (Main Page): The Figma plugin allows users to either type in the conversation manually or record audio directly during meetings. All recorded conversations are automatically saved in the “Old” tab, along with the original text and associated client information for future reference.
Keyword Extraction: After generating the resulf from the conversation, the next page will show several categories of design related field with the original sentences that include such label. The user can also generate an unique report that include all the suggestions from the conversation
Sentiment score: The sentiment socre will display as a bar graph with client ‘s attitude based on AI interpretation.
Client profile: The client profile library is an archive of past clients. Thie feature allows users to be more familiar with the client prefence when they decide to collaborate again.
Reason for changing to a web service
While implementing the design in the Figma plugin, I encountered challenges with accessing the device's microphone and persisting client profiles after the plugin tab was closed. These limitations made it difficult to maintain a seamless user experience. As a result, our team decided to transition the product to a standalone React.js application, which offers more flexibility for handling media input and storing user data persistently.
Version Two - Deployed Web Service
Testing ResultI then ran several tests for different design conversations and design categories.
Next Step
- Desktop Application
- Online Hosting Platform/Service