Numerical Computing Software for a Major Japanese Manufacturer

Challenge

A major Japanese manufacturer (listed on the Tokyo Stock Exchange) wanted to turn advanced numerical computing techniques into in-house engineering software for its engineers' daily work. The difficulty was that both ends had to be solid. On the algorithm side, it meant large-scale scientific computing with parallel acceleration. On the engineering side, it meant commercial-grade software that plant engineers could operate without a research background. And all communication with the client had to be in Japanese, directly, throughout.

Solution

Scope of work: The full software lifecycle, from numerical algorithm research to enterprise-grade delivery and two years of maintenance and iteration.

The project covered every stage of software development:

• Numerical algorithm design and implementation: Turning advanced numerical methods into solvers that run reliably at the scale and boundary conditions of industrial problems.
• Parallel performance optimization: Parallel acceleration and memory optimization of the core solver, bringing computations that used to mean long waits into a pace engineers can work with.
• Post-processing and visualization: Post-processing analysis and graphical display of results, so engineers can view results and make decisions right in the interface.
• GUI and enterprise packaging: An interface built for non-research staff, packaged as software that installs and runs directly in the client's internal environment.
• Direct, independent communication in Japanese: Chaired more than 20 monthly technical meetings and produced more than 30 technical reports and requirement documents, closing the loop from requirements definition and design review to acceptance and delivery.

Results

• 6x efficiency gain: Parallel optimization made the core solver about 6 times more efficient.
• Delivered and in use: The software has been delivered in-house and is operated directly by plant engineers.
• Nearly two years of work, and renewed: Continuous collaboration from 2024 to 2025, now moving into the next phase of feature expansion.
• Tech stack: C / C++ / Python (NumPy, PySide6, PyInstaller).

For manufacturers, this project proves one thing: every step from algorithm research to software that installs on an engineer's computer can be done by the same team.