WMS+TMS CHATBOT AND AI PROJECT

By

Bassel Matta,

Cigdem Polat,

Chaopin Wen,

Marco Ma

​

​

Project Goals and Objectives

1. Enhancing the existing Unis Live Chat through the transition to a Language Model (LLM) Chatbot trained with historical customer email and Jira case data.

2. Upgrading the existing WMS order process flow by implementing AI algorithms, such as OCR, to automate jira tickets creation by extracting order details and optimize the order creation process. This automation will eliminate the need for manual data entry, resulting in improved operational efficiency, cost savings, and error reduction.

2.1 – Automate Jira tickets creation for order creation

2.2 – Automate Jira tickets creation for LT claims

2.3 – Automate Jira tickets creation for routing

​

Current System Overview

Unis Live Chat

• Current Live Chat runs on www.unisco.com .
• Current system is integrated with Rocketchat, Amazon Lex and Unis APIs.
• When a customer, carrier, or potential driver initiates a chat, they will be redirected to live chat, inquiry forms, or website links based on their specific request and in accordance with the business logic.

WMS Order Process

• Inbound/Outbound orders are taken via EDI/FTP and email (customer.facility@unisco.com).
• When an email is submitted, a Jira ticket is automatically created. When the Data Entry Team receives the ticket, they manually create the order in WMS.

​

​

Project objective

Unis Live Chat

Enhancing Unis Live Chat Support through AI-Powered Automation

• Phase 1 (Completed):
• Utilize the Amazon Lex engine to establish the basic architecture and flow of the Chatbot. Currently live on Unisco.com.
• Utilize the newly implemented Rocketchat server to connect customers with support agents.
• Resolve the most frequently asked questions from customer by providing clickable paths that lead to relevant answers.
• Provide answer through website links for direct access.
• Utilize the existing TMS public API to provide customers with the location of their queried shipment
• Request customers to fill out an email form and send it to the corresponding customer service representative

​

​

Phase 2:

1. (Action 1) Further enhance Amazon Lex language model training and tuning
2. (Action 2) Email and Jira tickets data collection
3. (Action 3) Training and fine-tuning a language model using TMS and WMS’s historical email and Jira ticket.
4. (Action 4) Open a TMS and WMS quotation public API that is accessible through live chat.
5. (Action 5) Implement agent and customer queue and waiting list in on Rocketchat.
6. Further categorize Rocket chat system into multiple departments and incorporate Philippines support team.
7. Integrate with 3CX phone system.

​

Action Plan 1.

Training Amazon Lex

• Build intents and Utterances: Define the intents and create representative utterances based on the collected training data. Map the intents to corresponding actions or responses to ensure accurate and meaningful interactions with the customers.
• Define Slots: Identify and define the necessary slots or parameters that are relevant to customer queries. Configure the slots to capture specific information required for accurate intent and response.
• Testing and Evaluation: Test the trained Amazon Lex bot using sample customer queries and evaluate its performance.

​

​

Responsible Party: Chaopin

Action Plan 2.

Data Collection

Collect Data: Gather a diverse and representative dataset of customer support interactions, including customer queries and corresponding support team responses from emails (cs@unisco.com) and Jira tickets (ops.logisticteam.com and jira.logisticteam.com)

Data Preprocessing: Clean and preprocess the collected email and Jira ticket data. Remove irrelevant information, such as email signatures or system-generated content, and focus on the core customer queries and support team responses. Standardize the data format for consistent analysis.

​

​

​

​

Responsible Party: Bassel, Chaopin

Action Plan 3.

Training and Fine-tuning a Language Model

• Data Annotation: Develop an annotation schema to label the data with relevant information, such as customer queries, support team responses, intent, resolution, and any additional data. Annotate the data based on the defined schema to prepare it for training and fine-tuning.
• Model Selection and Setup: Choose a suitable language model architecture, such as BERT, GPT, or Transformer-based models, that supports training and fine-tuning on custom datasets. Set up the selected model and configure it for training and fine-tuning using the annotated dataset.
• Fine-tuning: Fine-tune the language model using the annotated customer support data. Adjust the learning rate, batch size, and other hyperparameters to optimize the fine-tuning process.

​

Responsible Party: Bryan, Reck, Foster

Action Plan 4.

Open a TMS and WMS quotation public API

• Requirement Analysis : Understand the specific requirements for opening a TMS and WMS quotation public API. Determine the necessary functionalities, data inputs, and outputs required for the API integration with Unis Live Chat system.
• API Design: Define the API design, including the endpoints, data structures, request/response formats, and authentication mechanisms.
• Integration with Unis Live Chat: Identify the integration points between the API and the Unis Live Chat. Ensure data exchange and compatibility between the Unis Live Chat and the API.

​

​

Responsible Party: Bryan, Quinn

Action Plan 5.

Implement Agent and Customer Queue on Rocketchat

Queue Design : Define the design of the agent and customer queues within Rocketchat, taking into account of the Rocket Chat API. Determine the queue structure, priority handling, notifications, and routing logic within the Rocketchat API.

Setup an Amazon Lambda API: Create an Amazon Lambda function to act as an API endpoint for handling queue operations. Set up the necessary permissions and configure the Lambda function to respond to the required API requests.

Integrate with Live Chat System: Identify the integration points of Unis Live Chat. Modifying the existing Live Chat to communicate with the Rocket Chat API for queue management.

​

​

Responsible Party: Bryan

Project objective

WMS Data Entry Process

Automating the WMS Data Entry Process for Improved Efficiency and Accuracy

• Phase 1:
• Evaluate the existing data entry process in the WMS system to understand the specific areas that needs the improvement. Identify the types of documents involved (PDF or Image), the fields that require data extraction, and any existing challenges. Define the requirements for the OCR algorithm and document processing system. Customize the OCR solution to match the specific data fields and document formats used in the WMS system, ensuring accurate extraction and integration.

​

Action Plan 1

Develop OCR Algorithm

• Evaluate the Data Extraction Requirements: Conduct a thorough evaluation of the data fields and information contained in the Jira tickets that need to be extracted and transferred to the WMS system. Identify the specific data elements, such as order numbers, order descriptions, customer details, and relevant timestamps.
• Define the Data Mapping: Determine how the extracted data from the Jira tickets should be mapped and integrated into the corresponding fields within the WMS system. Ensure compatibility and consistency between the data structures of both systems.
• OCR Implementation: Integrate OCR technology into the data extraction process. Train the OCR algorithm using a diverse set of Jira ticket samples to recognize and extract the relevant data fields automatically. Configure the OCR system to accurately capture text from the Jira ticket images or PDFs.

New CHATBOT LOGIC

New ocr SYSTEM

CURRENT SYSTEM (WISE)