Generative AI web app to automate business operations
The challenge
We needed to create a solution that would allow the client to optimize document search, email classification, customer support, and query categorization. The main goals were to reduce the use of human resources and cut expenditures and time on routine business operations.
Delivered value
We developed a GenAI web app for document search, email classification, customer support, and query categorization. The app reduced human error within processing data and customer requests. Our solution minimizes the processing time, provides faster response times, and increases user satisfaction with the client’s service. Moreover, the app improved request processing for various customer groups due to integrated accessibility standards and worked effectively for non-native speakers.
The process
To help our client automate routine information processing operations, we created a web app with four GenAI solutions: quick document search, email classification, customer support chatbot, and support request categorization.
You can book the web app demo to see how our solutions work. Our team can help you implement any of the four GenAI features listed below:
1. GenAI document search
The document search tool streamlines the process of finding specific information within a custom dataset, such as company policies, legal documents, or contracts. The model comprehends document context to deliver precise answers tailored to the context of the user’s question and relevant keywords.
Our data science engineers used the Llama index, a data framework that connects large language models (LLMs) and a client’s custom data set, and an OpenAI model. Llama index works as a query engine, capable of providing single question-and-answer interactions, and can be upgraded to a chat engine. This setup ensures the client’s sensitive data is securely indexed, forming a searchable knowledge base specific to their domain.
2. Email classification
This NLP solution was configured to a client’s business context to facilitate email classification and prioritization since it is done automatically.
To train the algorithm, our data science engineers used an open dataset comprising 20,000 emails categorized into 20 distinct classes and subclasses. This is a single-class classification, which is effective for anomaly detection (unusual patterns in the text), sentiment evaluation (understanding the emotional background of the text: positive, negative, and neutral), user intent estimation (realizing the user’s intent), and pinpointing specific events within a text.
Our team of data science experts used the TF-IDF algorithm to extract relevant features from text data for effective distinction between different classes. Employing a Naive Bayes classifier for multinomial models, they classified emails into classes and subclasses for efficient management and analysis.
3. FAQ chatbot
This GenAI solution helps to establish prompt and personalized interactions with a client’s end users. Even if the volume of inquiries and responses increases, an NLP FAQ bot maintains efficacy and speed.
To develop an algorithm, our data science team leveraged the OpenAI embedding model, TF-IDF algorithm, and Faiss model. The Faiss model helps the GenAI algorithm to search through a large database of questions and answers in real time and allows for FAQ customization. The OpenAI embedding model allows the algorithm to grasp the meaning of user queries and match them to relevant answers in the database. The TF-IDF algorithm helps the algorithm to understand the context of the queries, which ensures that the chatbot generates tailored responses.
4. Classification of queries into classes and subclasses
Our data science engineers enriched the web app with a GenAI solution that automatically labels and categorizes queries. These sorted inquiries can swiftly be conveyed to the respective departments for subsequent processing.
To accomplish this classification task, we used the ChatGPT model since it understands the context of requests. The algorithm undertakes primary and secondary categorizations based on the question context. It can grasp different language patterns and figure out user’s intentions even if they are ambiguous.
We are looking forward to you trying our GenAI solutions out and integrating them into your system. You can provide your custom data set to train the algorithms and check how they perform in your specific domain.
How it works
