Generative AI has garnered significant attention in recent months, captivating people with the remarkable level of conversation they can engage in with ChatGPT. Simultaneously, professionals are actively seeking ways to harness this tool’s potential. Many blog posts and videos abound, asserting that generative AI is a game changer poised to revolutionize numerous domains while introducing novel services and approaches.
The rapid pace at which this technology advances has instilled a sense of urgency, as nobody wants to miss out on the opportunities it presents. The ease of use and versatile range of applications contribute to its swift adoption, positioning it as a multipurpose tool.
People quickly realized that ChatGPT 3.5 and the underlying generative AI model have limitations in answering questions related to recent months, given its training cut-off in 2021.
However, professionals are eager for a generative AI system to comprehend company-specific data. Fortunately, we are lucky, as generative AI is a versatile tool. It offers the opportunity to incorporate custom data, enabling it to provide more accurate answers and perform tasks relevant to company-specific information.
Two main methods to incorporate custom data are:
- Prompt engineering
- Large Language Models fine-tuning
While both methods can be useful simultaneously, each has advantages and disadvantages for the particular use case. The rest of the article will focus on the prompt engineering option. Various generative AI models can be of use. We are using GPT-3.5 Turbo as part of the solution.
In numerous scenarios, we seek to integrate generative AI into our processes, allowing it to answer queries, gather and analyze data, create steps to achieve a goal, or perform specific tasks. Most of these tasks require that generative AI is “aware “of the context. To achieve this, we can turn to a concept that excels in situational awareness—the digital twin.
Digital twins and generative AI
The digital twin concept offers numerous features that we can leverage. This article will focus on refining the digital twin to function as a system of systems with situational awareness. Situational awareness software built on top of the digital twin concept provides data that reflect the current state. For our exploration implementation, we have chosen the fleet management domain. Our simplified fleet management digital twin model takes the following form:
The digital twins graph for three vehicles and a couple of destinations looks like this:
In addition to its various applications, the primary advantage of generative AI lies in its ability to facilitate human-like interactions with different users. In the context of our fleet management solution, these users include the fleet operator, driver, and customer. Each interaction necessitates real-time contextual information by retrieving data from the digital twin graph. There is no reason to use technology without business benefits:
- Customer-oriented use cases
- reducing customer service costs
- introducing new services
- enhancing customer experience
- improving customer satisfaction
- Driver-oriented use cases
- increasing driver productivity
- improving interaction with customers
- Fleet operator-oriented use cases
- increasing fleet operator productivity
- extending application usability
Let’s delve deeper into particular use cases.
Customer use case – delivery is late
Location data from vehicles undergo processing via standard IoT pipelines, and the vehicle position samples get directed to the digital twin service. When a vehicle’s digital twin receives a position update, it triggers the estimated arrival time (ETA) calculation for scheduled destinations. After a few position updates, the business and decision logic detects that the vehicle is moving slower than anticipated (e.g., due to a traffic jam), resulting in an ETA that exceeds the delivery window communicated to the customer earlier in the day.
Depending on the customer’s status, we will automatically contact the customer to provide updated information regarding the actual state of affairs. We will inquire whether the customer will wait, cancel, or reschedule the delivery. This particular task aligns well with the capabilities of generative AI. Our responsibility is to construct prompts instructing the generative AI on how it should behave and furnishing it with relevant contextual information.
In this case, the generation of prompts is from three different sources:
- Digital Twin Graph: current delay, contact to the driver, the reason for the delay, customer status
- Delivery scheduling service: provide information about available time slots for rescheduling
- Company data: Information about the benefits of platinum, premium, and standard customer statuses.
Once the chat session completes, generative AI is utilized for a second time to assess the conversation and deliver the customer’s final decision. Following that, employing additional services helps to execute the customer’s chosen course of action and make necessary adjustments to the state of the digital twin.
As we observe from the prompt composition, it consists of data sourced from multiple channels. Static data encompasses general information about customer status types and associated company policies that do not undergo frequent updates. Adding the entire dataset in the prompt might be unnecessary since it is extensive, but it is vital to incorporate the customer’s query with relevant data. Employing text embedding indexed in a vector database is a suitable solution in this scenario.
The digital twin graph serves as a valuable source of real-time contextual data. In this customer-oriented use case, a predefined set of digital twin graph queries retrieves all the relevant data. By employing these fixed queries, we ensure that the user receives only the customer-specific information, eliminating the need for a separate filtering or data security layer.
The scheduling service offers a list of available rescheduling slots and verifies the validity of the chosen time slot.
Driver use case – be aware of changes.
The primary responsibility of a driver is to safely operate the vehicle, execute planned tasks, and provide a positive customer experience. The interaction between the driver and the customer is often the only personal contact between a company representative and the customer. They require support to alleviate the workload and ensure they can fulfill their responsibilities effectively.
The service execution or delivery process is prone to changes. For instance, rescheduling, altering the contact person’s phone number (where a relative will pick up the package), or changing the delivery point and contact person (where a neighbor will handle the package) are common occurrences. The process described in previous customer-oriented use cases is useful in managing such changes. The digital twin graph captures these changes.
The interaction between the driver and the generative AI system is voice-based. The driver is informed about upcoming stops and can receive contextual information to adapt to changes. This approach presents an opportunity to brief the driver with customer-related data, thus enhancing the overall customer experience.
Regarding implementation, most of the data originates from the digital twin graph. However, there may be instances where the driver has additional questions that necessitate modifying a graph query. We use generative AI to directly translate the driver’s question into a graph query to address this. A data security layer is required to ensure data security and filter queries beyond the information boundary.
Fleet operator use case – be aware of the current state.
The fleet management solution encompasses business and decision logic, which monitors operational processes and initiates alerts or automated actions. The fleet operator notifies the outcomes through a management dashboard. The fleet operator can benefit from generative AI by posing questions translated into graph queries.
The query results can provide customized answers specific to the current situation beyond what the dashboard captures.
Generative AI can effectively assist the fleet operator in achieving a goal to address the current fleet operation state, and breaking the goal into specific actions help in the execution of successful attainment. This assistance ensures that operators recognize all crucial steps and facilitate the discovery of novel approaches to accomplish them. Acting as a copilot, the generative AI offers suggestions and can carry out tasks as an agent.
We have successfully incorporated a customer use case into our fleet operation exploration solution. By extracting contextual information from a digital twin graph and leveraging generative AI, we have significantly enhanced the capabilities of our solution. It enables each role in our schema to access up-to-date information and make informed decisions. The generative AI serves as a human language interface, information extractor, and query transformer.
Unleash the true potential of your business with the dynamic combination of digital twin and generative AI. Seamlessly connect the contextual insights from your digital twin to the creative power of generative AI. Gain a competitive advantage by leveraging real-time data and predictive capabilities from your digital twin, amplified by the innovative possibilities of generative AI.
Make informed decisions, streamline operations, and unlock untapped growth opportunities with unparalleled efficiency and optimization. Experience the future of business transformation today with our cutting-edge digital twin integrated with generative AI.