AI-Powered Ticketing System Transformation in Transportation Construction

# Data Analytics # Web development
Discover more

Highlights

  • Delivered AI Expertise for Construction Industry: We bolstered the transportation construction system by integrating AI-powered functionalities such as image processing, report generation, data classification, traffic analysis, and chatbots.
  • Utilized Multiple AI Models: The project involved leveraging a diverse range of AI models, which were effectively employed through the use of client data and targeted prompts. This approach ensured the delivery of accurate AI-generated responses. 
  • Ongoing Collaboration with HaulHub: Following the successful completion of our previous project, the client re-engaged with us to leverage our AI expertise in enhancing their product.

Client

HaulHub is a B2B2C company that provides a digital platform for the transportation construction industry. Its focus lies in leveraging artificial intelligence by carefully collecting, securely storing, and strategically utilizing AI technologies and models. We have a longstanding and continuous partnership with HaulHub. Previously, we developed web and mobile applications for business intelligence and established a robust data processing solution capable of efficiently managing millions of analytical data points. Additionally, we implemented a scalable solution for the company using serverless architecture.

The company has established partnerships with over 35 state agencies and more than 500 contractors in the US. HaulHub facilitates the transition from traditional paper-based methods to AI-powered solutions in order to digitize the construction material supply chain. This shift involves moving away from outdated paper-based systems and towards a centralized digital repository. The goal is not only to reduce project delays, safety risks, and inefficiencies associated with manual tracking, but also to improve the speed, safety, and efficiency of construction workflows.

Website:
Country:
Industry:
Team Size:

Product

The product is a web and mobile solution that provides a ticketing system for collecting data on construction materials, including asphalt, concrete, and sand. It then analyzes this data for customers, which could include current or different plants, as well as external companies involved in road construction. It’s specifically tailored for use by both plants and transportation departments.

The company has several internal products for analyzing data related to road, bridge, and highway construction. These products track details such as who worked on the project, when it was completed, and then perform additional analyses on top of this information.

Goals and objectives

Our main goals were to integrate AI the product for the following purposes:

  1. Image Processing: Implement AI to analyze construction photos, ensuring people safety and identifying areas for process improvement.
  2. Report Generation: Utilize AI to analyze daily work reports and generate concise summaries tailored to clients’ specific requirements.
  3. Data Extraction and Classification: Employ AI to analyze documents, extracting and categorizing data according to specified criteria. This analysis serves internal needs within the company.
  4. Traffic Analysis: Leverage AI to analyze traffic patterns across construction areas, providing insights into how construction activities impact road traffic before, during, and after construction.
  5. Chatbots: Develop chatbots to facilitate conversations between clients and the data layer using AI. The AI determines which data to fetch from the HaulHub database, prepares it for the client, and responds accordingly. This process unfolds in a conversational mode, akin to natural dialogue.

Project challenge

  1. Development and Training of AI Models: The most challenging aspect of the project was training AI models, especially for ensuring people safety and traffic analysis. This was due to the lengthy process of comprehending the nuances of business requirements and converting them into a format that AI could understand. Extensive research and manual AI requests were essential to accurately define our capabilities and limitations.
  2. Data Analysis: AI needed to take big pieces of data, analyze them fast to get the most important facts that could be used to improve the construction process.

Solution

We used AWS Lambda as the foundation for our client-facing APIs. These APIs enabled clients to make requests, which were then processed by AI services to produce results based on the provided data. We employed various AI models and evaluated their effectiveness across different tasks. These models were:

  • OpenAI ChatGPT 3.5 and 4: Applied to tasks involving natural language processing, such as generating responses to user queries, facilitating conversations, and understanding textual data in context.
  • AWS Bedrock Claude: Utilized for tasks like summarization and document comparison, particularly for drafting reports.
  • AWS Bedrock Titan: Applied for image, multimodal, and text models, enabling a wide array of generative AI applications including content generation, image processing, and search and recommendation systems.
  • AWS Bedrock Jurassic: Employed for sophisticated text generation tasks in enterprise settings, such as question answering, text generation, and summarization.

We relied on HaulHub’s data and detailed prompts to ensure accurate AI responses. Data quality assessment involved both human evaluators and algorithmic measures. Additionally, we imposed strict limitations on AI usage before each request and thoroughly analyzed the responses.

Now, we are integrating a RAG (Retrieval-Augmented Generation) approach for chatbot conversations which helps with providing relevant context information and ensures fluency in generated responses. According to this method, the retriever component extracts relevant information from the HaulHub data sources, while the generator takes the retrieved passages as input along with the original query and generates an accurate, quick and context-driven response. 

Tech Stack

  • Java Java
  • Quarkus Quarkus
  • React.js React.js
  • Ruby Ruby
  • OpenAI OpenAI
  • AWS Bedrock  AWS Bedrock

Our results

We effectively integrated several AI models into the system, including OpenAI ChatGPT, AWS Bedrock Claude, AWS Bedrock Titan, and AWS Bedrock Jurassic. These models serve purposes such as image processing, report generation, data extraction and classification, traffic analysis, and chatbot functionalities. Our efforts to enhance this process are ongoing, and we maintain close collaboration with the client to ensure project delivery.

Next Project
# AWS infrastructure, # Data Analytics, # Fintech, # Web development
Developing a Web Crawler Application for One of the Premier U.S. Asset Management Companies

Highlights Client Our client is one of the premier U.S. asset management companies that employs more than 12 000...

Explore Case