Faced with a long-standing product traceability challenge at a US automaker's battery manufacturing plant, Mr. Le Anh Khoi developed the first version of a solution in just four weeks with the support of AI.
A Complex Challenge
Three months ago, Mr. Le Anh Khoi of FPT Automotive in the US, FPT Corporation, began developing a new solution for the quality management department at a partner's battery manufacturing plant.
The requirement was to establish end-to-end product traceability, enabling the manufacturer to track the entire journey of a battery, from raw materials, components, and production batches to every manufacturing process involved in creating the finished product.
This was not a new challenge. Several teams and international partners of the automaker had previously developed tools to address the problem, but none had fully met expectations.
The difficulty stemmed from the complexity of battery manufacturing data. A finished battery consists of multiple battery cells, raw materials, components, and intermediate products. When a quality issue arises, the manufacturer must trace the entire chain of relationships to identify the root cause and determine the scope of affected products.
These data relationships form a complex "family tree" involving hundreds of interconnected elements, making traditional relational database models difficult to scale and inefficient to query.
Mr. Le Anh Khoi of FPT Automotive in the US, FPT Corporation.During his research, the FPT engineer realized that the biggest obstacle was not the user interface or data visualization features, but the underlying data architecture.
Instead of continuing with a traditional relational database, he selected Neo4j, a graph database platform well suited to tree-like data structures and complex networks of relationships.
Manufacturing data was transformed from flat tables into a graph-based model. This enabled the system to visualize relationships among finished products, battery cells, raw materials, production processes, and related batches.
Users could not only trace the origin of a battery but also examine the entire manufacturing journey through an intuitive visual diagram and export data for further analysis and quality management.
According to Mr. Khoi, this capability provides a critical foundation for the manufacturer to accurately determine which products need to be held, which batches require additional monitoring, and which areas of risk must be controlled.
AI Accelerates the Journey from Idea to Implementation
Mr. Khoi revealed that one of the most remarkable aspects of the project was that most of the application's source code was generated with AI assistance.
Rather than focusing on writing detailed prompts for AI to generate code, he spent most of his time defining requirements, guiding the design, reviewing the architecture, validating results, and continuously refining AI-generated outputs.
Within approximately four weeks, the first version was ready for presentation to the customer.
After experiencing the core features, the partner quickly recognized the solution's value and proposed additional requirements, expressing interest in accelerating deployment across other battery product lines.
However, AI was not where the story began.
Mr. Khoi explained that the FPT team had been involved in the project from the early stages, providing Application Management Support (AMS). Through this experience and the use of GitHub Copilot, the team developed an in-depth understanding of the system and identified an unmet need.
“We recognized an opportunity to develop a new approach to product traceability,” Mr. Khoi shared.
The project was originally developed using Siemens Opcenter and Mendix, low-code platforms designed for manufacturing environments.
According to Mr. Khoi, these platforms also presented limitations in terms of AI capabilities. AI performs best when it has a clear understanding of the framework, system architecture, and technical context. With platforms containing extensive customization or proprietary data, AI often lacks sufficient information to generate optimal recommendations.
“If a platform does not adequately support AI, or if AI cannot understand the platform's data, it becomes very difficult to unlock AI's full potential,” the FPT engineer observed.
Based on this insight, he decided to build an independent system using a traditional software development approach, allowing AI to participate directly in source code generation, frontend development using plain JavaScript, and the implementation of business requirements.
Mr. Khoi understood that while AI can generate large volumes of code in a short time, it cannot yet guarantee architectural correctness or product quality.
Therefore, rather than concentrating solely on code generation, he devoted considerable time to code reviews, testing, and system architecture assessments.
He also frequently asked AI to generate accompanying unit tests to verify code quality from the outset.
In his view, as AI becomes increasingly involved in software development, the essential skills of an engineer will extend beyond writing code to include reading and reviewing code, evaluating system design, and identifying risks that AI may overlook. Mr. Khoi also believes there is no such thing as a “magic prompt.”
Today's AI tools are powerful enough to generate multiple possible solutions. What makes the difference is not the complexity of the prompt, but how well users understand the problem, the objectives, and the limitations of the system.
Without a clear understanding of the problem to be solved, even the most carefully optimized prompts are unlikely to deliver meaningful value.
Conversely, when engineers fully understand business requirements and customer objectives, even simple prompts can produce effective results.
Reflecting on the journey, Mr. Khoi emphasized that while AI helped turn ideas into reality much faster, identifying the right problem and defining what needed to be solved remained fundamentally human responsibilities. “What I am most proud of is not that we had an advantage from the beginning, but our ability to learn very quickly and our willingness to take on challenges,” he said.
And although AI is increasingly capable of handling a broader range of tasks, ultimate accountability still rests with the developer.
“AI can write code incredibly fast, but people must remain responsible for understanding, reviewing, and ensuring the quality of the final product,” Mr. Khoi concluded.