top of page
Company Overview
Griffith Foods is a family-owned company that develops and manufactures high-quality food ingredients. As a global product development partner, they are committed to aiding their customers in creating superior products, their focus extends to building a better, more
sustainable world.
Griffith Foods faces a critical challenge in efficiently accessing essential maintenance records and data for heavy machinery across their global factories for troubleshooting. This challenge is further compounded by intricate language, time zone disparities, and diverse cultural contexts prevalent across their global footprint.
Problem Focus
"Prior to the project; Griffith Foods had identified the need for an LLM-enabled expert system (AI) and wanted to leverage cutting edge technologies to maintain a competitive edge".
Acknowledgement
Problem Statement
"How to seamlessly integrate LLM into Griffith's daily operations, leveraging the specialized knowledge and hands-on experience of employees with floor equipment and activities?"
Solution Space
"Development of an LLM-enabled expert system (AI) for the purpose of querying and answering questions relating to the systems, practices, records, and operations within the organization".
Project Goals and Expectations
Analysis
Understand the impacts of an NLP system.
Delivery
Delivering a Minimal Viable Product (MVP).
Research
Evaluate the current needs of Griffith Foods employees. Research current NLP systems in market.
Project Timeline
Understanding users
My next step was to understand how our user behaves. Partnering for research with Griffith Foods; we identified that our Primary users are-Plant Maintenance workers, Engineers, Mechanics and Supervisors.
User Archetypes over Personas- Focus on Actions Over Demographics
Efficiency Enthusiast
Behaviors:
Embodies a relentless pursuit of optimization in maintenance. With analytical precision- sets ambitious goals, leveraging technology and data to drive efficiency.
Goals
Focuses on four key goals: enhancing equipment efficiency, improving troubleshooting procedures, integrating smart manufacturing processes, and upgrading equipment.
Frustrations
Troubleshooting equipment on the get go, Inadequate staff training, communication gaps in maintenance planning.
Needs
Wants an answer quick. Going through a 2000 Page manual for a minor troubleshoot doesn't make sense. Wants to leverage Griffith's massive maintenance database to formulate answers.
"Calling a Maintenance team consumes time!"
Goals Defined
User and Client Expectations
The system should be able to process queries or requests raised by users and allow for uploading new information.
The system should be able to provide the users with an accumulated answer from the various data sources available with the system.
The LLM enabled system should be able to synthesize and tag the text documents such as training and manuals of engineering equipment.
Information Architecture
Simplified user flow (Search Data)
User inputs query into NLP system
System generates search results based on ALL data uploaded in system database
User selects required channel of response (Video, Images, Docs etc).
Simplified user flow (Upload Information)
User chooses to upload new information onto the NLP database.
The user has to describe what information they are uploading.
User can upload related data and choose the type of data being uploaded.
Process
Wireframing
-
A search bar to input user query.
-
Filter options to narrow down search results and channels for ease.
-
As mentioned, filters will assist user's search query.
-
All types of results relevant to the query will be displayed.
-
The system asks user if the query has been resolved. If not the user has an option to create a personalized ticket.
Coded prototype to understand the functionality better
-
The prototype was coded in HTML and Java Script. Material.io libraries were imported for UI elements.
-
An OpenAI key was downloaded to test the NLP's capabilities to browse databases.
Final Designs
User can search for data based on relevance- Maintenance, Processes & Equipment. This will alter the results the user sees
Upload information- Databases, documents, images etc.
User can change the facility they are accessing the terminal from.
-
Based on search query, user is provided with different channels (Generated text, Images, Links, Docs etc.) of information for assistance.
-
The user can access the provided information.
-
If the user isn't satisfied with the given information; they can raise a ticket for a more customized approached.
User can choose the category of file to upload into database.
Provide a description for the file(s).
Upload required files (Of any type).
-
Filters direct user towards relevant results.
-
In the LLM model, Documents, images, videos etc are tagged with their respective filters.
Design System
Potential Impacts of NLP
Implementing an LLM-enabled expert system will allow information retrieval within the organization, seamless querying and answering of questions regarding systems, practices, records, and operations.
NLP can enhance quality control and process improvement at Griffith Foods US by analyzing data for patterns and trends, automating quality checks, and identifying areas for improvement in production processes.
NLP serves as an employee database, retrieving essential information based on specific skills, and issuing alerts based on performance metrics and machine statistics to enhance plant efficiency.
bottom of page