Since the public debut of OpenAI’s ChatGPT in November 2022, the emergence of Generative AI has sparked the creation of numerous startups.
But is Generative AI a buzz, or is it a real deal?
As an AI technology professional, I embarked on a mission to explore the potential of AI by developing an internal-use knowledge base chatbot using Teams and Azure OpenAI. The goal was to empower each Sales Talent with their own Digital Assistant, in order to keep our Talents up-to-date with our ever-expanding product and service portfolios at HKBN.
Through this month-long experiment, I came to the realization that training ChatGPT is like training my two little kids at home. To fully harness the potential of ChatGPT, I divided the training process into multiple stages and assigned dedicated GPT agents to each stage. By dividing the workload among GPT agents, I ensure each stage receives focused and specialized attention from the GPT agent that is tuned to handle particular task, thus leading to optimized performance for the chatbot. This is very similar to how I separately teach different subjects to my kids based on their academic aptitude, so that they can excel in their respective areas of expertise.
This article delves into the four key challenges encountered during the development of HKBNGPT and the innovative solutions employed to overcome them.
 |  |
| Distinguishing between questions requiring answers from internal or public data | Ensuring classified information remains confidential when interacting with the chatbot |
One of the primary challenges in building a chatbot is determining whether a question requires an answer derived from internal or public data. To ensure accurate responses, understanding the context of each question is crucial. I tackled this challenge by implementing an additional classification layer in the system. This layer allows the chatbot to identify if a question pertains to internal data by cross-referencing with the available internal data, enabling HKBNGPT autonomously determine the relevance of a question to internal data.
Challenge 2: Addressing ambiguous questions
Ambiguous questions pose a challenge in extracting relevant internal data to provide meaningful responses. To overcome this issue, I incorporated an extra layer of prompts using GPT technology. These prompts enhance the initial inquiry by incorporating relevant keywords and fine-tuning the question’s precision. This approach aids in efficient searching of the internal database and ensures accurate information retrieval. For instance, by enhancing a question like “How can HKBNES support SMEs?” with additional context and relevant details, the chatbot can provide more effective results.
Challenge 3: Identifying relevant internal data for a given answer
To identify relevant internal data for an answer, we employed a vector database based on semantic embedding similarity. By creating a vector representation of internal data, we can match the user’s question with the most pertinent information. Leveraging a pre-trained model for semantic similarity, HKBNGPT can compare the user’s question with the vector representation of internal data. This approach facilitates the discovery of the most relevant internal data, enabling the chatbot to provide personalized responses.
Challenge 4: Ensuring classified information remains confidential when interacting with the chatbot
Maintaining the confidentiality of classified information is imperative when developing a chatbot that utilizes internal data. To address this concern, an additional content filtering layer was implemented to filter out sensitive information before forwarding the question to GPT. This protective measure ensures that only relevant data is shared with the chatbot, upholding the privacy of internal information and preventing any leakage of classified data.
Introducing HKBNGPT: Revolutionizing Communication and Access to Knowledge
HKBNES is excited to introduce our enterprise-grade HKBNGPT Jumpstart Solution, which revolutionizes how businesses facilitate communication and provide access to their knowledge base for employees. By leveraging OpenAI’s advanced GPT technology securely deployed on Microsoft Azure, our solution guarantees the highest levels of data security. More importantly, it empowers users to access and enhance enterprise-specific knowledge.
Some of the key features of HKBNGPT Jumpstart include:
- Enterprise KnowledgeBase: A comprehensive repository of company-specific information.
- Prompt Management & Monitoring: Efficient management and monitoring of prompts to ensure optimal performance.
- Administration Portal for Knowledge Management: A centralized portal for administrators to manage and update the knowledge base.
- Role-based Query: Tailored access to information based on user roles and responsibilities.
- Public Internet Query: Integration with public data sources to provide broader insights.
By adopting this transformative approach, HKBNGPT Jumpstart drastically reduces the time spent on information retrieval, enhances workflow efficiency, and ensures all employees have accurate, up-to-date guidance at their fingertips. Our solution provides the utmost protection and reliability, enabling organizations to unlock the full potential of AI-powered chatbots.
In conclusion, HKBNGPT jumpstart solution represents a significant step forward in leveraging AI technology to address the challenges of expanding product portfolios and knowledge sharing within organizations. By employing innovative techniques to overcome challenges such as distinguishing between data sources, addressing ambiguity, identifying relevant information, and ensuring data confidentiality, HKBNGPT Jumpstart enables efficient communication and access to knowledge for employees. With enterprise-grade features and robust data security, our solution empowers organizations to streamline workflows, enhance productivity, and provide accurate guidance to employees.
Embracing the power of AI, HKBNES paves the way for a new era of intelligent and efficient knowledge management within enterprises.
