Streamlining Aircraft Payload Planning and Closeout with AI-Powered Chatbots on AWS

By Neeraj Kaushik, Executive IT Architect – IBM Consulting
By Adam Biener, Sr. Consultant and Architect – IBM Consulting
By Amit Chowdhury, Sr. Gen AI Partner Solutions Architect – AWS
By Ajit Kumar K.P., Sr. Gen AI Partner Solutions Architect – AWS

IBM

In the intricate world of airline operations, precise planning is the cornerstone of safety, efficiency, and cost-effectiveness. Among the myriad tasks ensuring smooth flights, the below-wing (aircraft operations handling on the ground) payload planning and the closeout process for aircraft is crucial. This process strategically arranges baggage, cargo, and fuel to optimize aircraft performance and adhere to stringent safety regulations.

IBM Consulting, with its extensive airline industry expertise, is positioned to address these issues by leveraging cutting-edge technologies on the AWS Cloud. IBM aims to streamline aircraft payload planning and closeout processes with Amazon Bedrock, a managed service offering foundational models (FMs) from leading AI companies, and capabilities for responsibly building AI applications with security and privacy in mind.

By reading this blogpost, you will learn how AWS generative AI services can be used to build applications that address end-user concerns in highly complex and regulated industries like airlines.

Traditional Process and Challenges

Payload planning is an essential process that enables airlines to optimize revenue while upholding safety standards. It involves meticulous planning of fuel allocation, bin and container positioning, and cargo considerations.

In a traditional payload planning process, ground agents develop load plans that strategically arrange baggage, cargo, and fuel within the aircraft. This iterative procedure adapts to dynamic changes in flight conditions, aircraft specifications, luggage volume, passenger numbers, and operational disruptions.

Ground agents must ensure compliance with Federal Aviation Administration (FAA) guidelines during planning and loading. The closeout process encompasses verification or audit checks to ensure the payload plan has been accurately implemented during aircraft loading.

These audit include:

• Position and Weight Audit: Confirm that baggage, cargo, and other items are positioned precisely as outlined in the payload plan to maintain proper weight distribution and balance.
• Bin Lock Audit: Verifying that all cargo bins and containers are securely locked in their designated positions.
• Freight Audit: Inspecting that all freight shipments have been loaded according to the plan.
• Special Handling Audit: Ensuring any items requiring special handling, such as live animals or hazardous materials, are accommodated per regulations and the payload plan.
• Destination Audit: Confirm that all items are loaded onto the correct aircraft for their intended destination.

Key challenges with the traditional processes:

1. During the payload planning stage, some common queries and issues faced by ground agents include:
   • Container Planning: Understanding container planning restrictions, such as destination mismatches, “Thru” containers, or containers with single, sharp, and heavy items.
   • Commodities/Containers Movement: Clarifying why certain commodities or containers were moved to a “no-ride” status.
2. During the loading stage, ground agents may encounter the following queries and issues:
   • Additional bags: Handling additional bags or containers not included in the original plan.
   • Handling animals: Determining the proper handling procedures for bags containing live animals.
   • Container placement: Understanding restrictions on container placement in specific forward or aft positions on the aircraft.
   • Unloading guidelines: Seeking guidance on unloading bags that need to be rerouted or returned to the claim area before flight closeout.
3. Inefficient processes:
   • Time-Consuming Adjustments: The placement of commodities or containers is frequently adjusted multiple times before the aircraft is cleared for takeoff by the weight and balance system.
   • Training and Optimization: Ground agents must be well-trained in all audits and procedures to optimize the closeout process and mitigate flight delays.

Solution Overview

The proposed solution addresses payload planning and closeout challenges by implementing an AI-powered chat assistant accessible through web and mobile applications. The assistant leverages generative AI and Large Language Models (LLMs) to provide real-time responses to user queries, overcoming the limitations of traditional statistical and rule-based approaches.

Figure 1 – Generative AI Solution for Payload Planning, Loading, and Closeout

To address these use cases, the solution leverages prompt engineering and the Retrieval Augmented Generation (RAG) architecture with real-time data access from various data sources to reduce the risk of hallucinations. As depicted in Figure 1, the solution ingests documents from various internal data sources into a vector database, by applying chunking and embeddings, to facilitate semantic similarity search corresponding to user queries. The solution also employs integrations with existing APIs to take actions or complete tasks in real-time based on user requests.

The solution will improve understanding of the existing system by leveraging code bases, training documents, and policy documents to respond to technical and functional queries from subject matter experts and the IT Team. Additionally, it will empower ground handlers with access to real-time information in the context of the flight.

Generative AI Solution on AWS

The solution leverages the capabilities of FMs and the other powerful tools provided by Amazon Bedrock to develop generative AI applications.

Figure 2 – Amazon Bedrock Solution for Payload Planning and Closeout

As depicted in Figure 2, curated content is uploaded to Amazon Simple Storage Service (Amazon S3). During pre-processing, Amazon Bedrock Knowledge Bases are used to ingest and segment these documents into manageable chunks. These chunks are converted into embeddings using an Amazon Bedrock embedding model, facilitating semantic analysis. The embeddings power a vector store index, enabling semantic similarity comparisons between user queries and customer data source text.

The user provides natural language queries through web applications or scanners, which are transformed into vectors using an Amazon Bedrock embedding model. Amazon Bedrock Agents use the preprocessing template to validate, contextualize, and categorize user input, interpreting it using conversation history, agent instructions, configuration, and the underlying Amazon Bedrock FM.

Amazon Bedrock Knowledge Bases offers fully managed RAG for Amazon Bedrock Agents to access customer data, configured by specifying usage instructions and linking to an Amazon S3 data source. Action groups, including planning, loading, and audit APIs, are defined in JSON files stored in Amazon S3.

During orchestration, Amazon Bedrock Agents utilize Reasoning and Acting (ReAct) prompting with the orchestration prompt template to complete the user’s task, incorporating action group API invocations and knowledge base queries to generate observations. These observations enhance the base prompt for the Amazon Bedrock FM, guiding the decision-making process. Using the knowledge base response generation prompt template, Amazon Bedrock Agents conduct semantic similarity searches to retrieve text and augment the base prompt with additional context.

FM Customization and Evaluation on Amazon Bedrock

Payload Planning, Loading, and Closeout is a specialized domain with unique terminology. Fine-tuning a model on a domain-specific corpus can enhance the model’s ability to understand and process the intricate details of payload planning, ultimately improving operational efficiency and safety. However, the solution could use fine-tuning to enhance its capabilities.

Amazon Bedrock supports fine-tuning and continued pre-training methods to customize Foundation Models (FMs) for specific use cases. Fine-tuning enhances model accuracy using task-specific labeled data, while continued pre-training improves domain-specific knowledge using unlabeled data. The continued pre-training method is particularly appealing for the unique terminology associated with the payload planning and closeout process.

After customizing your FM, you can evaluate its performance using Amazon Bedrock’s model evaluation jobs. These jobs support common tasks like text generation, classification, question answering, and summarization. For this use case, the question answering task can be used to evaluate the customized FM’s accuracy, robustness, and toxicity using built-in or custom datasets created with the help of domain experts.

Security on Amazon Bedrock

Handheld scanners and web applications used globally by airport personnel handle sensitive payload planning and loading data, including flight, baggage, and passenger information. Given the confidential nature of this data, robust security measures are crucial. Amazon Bedrock provides comprehensive security solutions including:

1. Data Encryption: Amazon Bedrock encrypts data at rest and in transit, using AWS-owned or customer-managed keys for encrypting custom models.
2. Guardrails: Amazon Bedrock Guardrails evaluate user inputs and model responses, allowing you to configure policies to avoid undesirable content and protect privacy.
3. Virtual Private Cloud: Using Amazon Virtual Private Cloud (Amazon VPC), you can protect enterprise data, monitor network traffic, and establish private connections to data sources.
4. Access Management: Amazon Bedrock access is managed using AWS Identity and Access Management (IAM), supporting identity-based and resource-based policies to define permissions.

Cloud security at AWS is the highest priority, and a shared responsibility between AWS and its customers. For more information refer Security in Amazon Bedrock.

Conclusion

In this blog post, the proposed AI-powered chatbot solution leveraging Amazon Bedrock and LLMs addresses the key challenges faced in the traditional aircraft payload planning and closeout process. By providing real-time responses to user queries, integrating with existing systems, and utilizing a curated knowledge base, the solution streamlines the process, reduces rework time and training requirements, and enhances agent productivity. This modernized approach helps ensure compliance with regulations, improves operational efficiency, and minimizes flight delays, ultimately benefiting both airlines and passengers.

Note, the architecture presented here is for reference purposes only. IBM will work closely with you to deploy the solution in accordance with industry standards and compliance requirements.

IBM — AWS Partner Spotlight

IBM Consulting is an AWS Premier Tier Services Partner that helps customers who use AWS to harness the power of innovation and drive their business transformation. They are recognized as a Global Systems Integrator (GSI) for 25 competencies, including Travel and Hospitality Consulting. For additional information, please contact an IBM Representative.

Contact IBM | Partner Overview | Case Studies

TAGS: AWS MSP Partner Program, AWS Partner Guest Post, AWS Partner References, AWS Partner Solutions Architects (SA), AWS Partner Success Stories, AWS Premier Tier Services Partners, AWS Public Sector Partners, AWS Service Delivery Partners, AWS Solution Provider Partners, AWS Well-Architected Partners, IBM, Managed Service Provider