Compass UOL leveraged generative artificial intelligence (GenAI) to develop a solution that automates part of this curation process by creating a script that automatically corrects spelling errors in chatbot dialogues while preserving the technical structure
The automated solution was designed to consider the data structure and its various specifics, such as environment variables and specific abbreviations. By applying GenAI and appropriate prompt engineering, the solution enables efficient text correction while maintaining the technical structure within the Watson workspace.
With this new solution, Sicredi’s curation team can now review and approve corrections more efficiently, leading to reduced time and cost for the process. This automation has streamlined the review process, making it faster and more effective.
The curation process for Sicredi’s employee-facing AI involves a series of challenges due to its complexity and the volume of information. This task required a detailed analysis of all dialogues entered into the platform, with each interaction needing review to identify and correct potential errors.
In addition to spelling corrections, curation requires specific attention to the use of variables and the integrity of technical aspects to ensure the AI functions properly for Sicredi. Given this scenario, there was a need for a solution that would not only speed up the process but also automate it.
Using Azure OpenAI (GPT-3.5 model), we developed an automated script designed to correct spelling errors in the conversational flows of the Theo chatbot, used for employee support at Sicredi.
The primary goal was to ensure linguistic accuracy in automated responses, preventing spelling mistakes that could compromise service quality.
The solution required meticulous handling of data structures, including environment variables, abbreviations, codes, and special characters, all essential for Theo’s proper functioning.
Performing spelling corrections
while maintaining the technical integrity of the workspace.
Enhancing the reading and correction process
by highlighting suggested corrections.
Establishing a continuous review cycle
promoting ongoing improvements to the system.
Reducing the time required for corrections and processing costs
resulting in operational efficiency.
Read other stories
PagueMenos
Brazil’s second largest pharmacy chain uses AI & ML to personalize product ranking and experiences
Compre & Alugue Agora
Brazil’s first and only social media for Real Estate reduces chatbot costs by 80% and improves responses with AI