Previous
Next
Client
Personal Project
Duration
Four Weeks
Role
Designer
Project Brief
My team and I were tasked with envisioning a novel service that employs NLP to co-create value for users and service providers. Using matchmaking and other evaluation tools, our goal was to rank AI concepts based on technical feasibility, financial viability, and consumer desirability, and choose one that is both high reward and low risk.
The Problem
Email scandals can lead to significant financial penalties for corporations, potentially resulting in legal battles that cost millions of dollars. Even with rigorous training protocols in place, there are instances where employees might resort to language that is considered offensive, whether through insults, discriminatory remarks or more. These scandals often attract public attention and can damage a company's reputation. This loss of trust can lead to a decline in business, loss of customers, and a drop in stock prices.
The Solution
DontSueMe.ai is an AI-powered product designed to create a safer workplace by utilizing email text analysis and classification. Our advanced Large Language Model analyzes the content of emails and identifies any hostile language, alerting users before they send out potentially harmful messages. This tool offers an additional layer of legal safeguarding for large corporations, potentially saving them millions in legal costs.
Why should companies adopt our product?
Email scandals are expensive.
Company reputation is affected by its employees.
Our product saves companies millions in legal fees.
Our product will identify potentially hostile keywords and issue a prompt notification upon detecting any content that could be deemed offensive or inappropriate. Users have the ability to either dismiss the alert or report the classification as erroneous.
Leveraging foundational models like GPT-4 and LaMDA, our "AI-Wrapper" product offers a cost-efficient development solution. By using only a minimal dataset for training and specialization, we substantially reduce the resources required. We also employ transfer learning techniques with these foundational models to enhance dataset labeling.
Sentiment analysis and hostile word detection are areas where NLP models already excel. To further test this hypothesis, we fed several email messages to GPT-4 and tasked it to detect hostility. GPT-4 was able to classify all instances correctly.
Our product benefits both employers and employees. We maintain a sustainable business model by charging employers a monthly premium, which in turn provides them with lawsuit protection. Employees gain an added layer of safety, receiving reminders before sending potentially inappropriate emails.
