AI-Charged: How We Cut Migration Planning Time by 70%

The product is a next-generation transportation management system that brings together data from warehouses, cars, and customer systems into a single control tower. But the true victory came from using AI-assisted analysis to speed up discovery and planning so that engineers could do their jobs more securely and quickly.

This project marked the first application of our acceleration framework, providing valuable insights into its effectiveness. Devox Software was recommended to this client. We discussed that our approach is pioneering; that’s why we had the client’s full confidence—and justified it.

The team included:

• A Solution Architect,
• .NET engineers,
• A DevOps expert,
• A QA specialist.

This way, every role was enhanced with AI-powered acceleration.

The challenges were much the same for the legacy modernization process. They included:

• lack of documentation,
• tight coupling of old services,
• secret rules,
• downtime-sensitive operations, and more.

However, our focus was on delivery timelines, since the first estimate for only manual finding and planning across three disconnected systems exceeded 10 weeks. The client couldn’t spend so much time. That’s why we sought ways to accelerate the process.

We utilized Azure DevOps with Terraform, containerized .NET 8 services, Angular for the admin interfaces, PostgreSQL as the target data store, and TensorFlow combined with Azure Cognitive Services to power automated code clustering, provide data lineage hints, and enable risk scoring within the AI Solution Accelerator™.

We set up the job as a pipeline based on evidence, reinforced by AI Solution Acceleration to any degree:

• Deterministic Discovery: We searched repositories and SQL workloads to figure out call graphs, shared schemas, batch windows, and other hotspots that weren’t evident. The result was a live system map with levels of confidence and gaps marked for interviews. 
• Set Boundaries. We created a small, versioned API known as the “contract of record” for managing orders, carriers, and tracking. This technique allowed teams to change things behind the scenes.
• Wave Planning Simulations. The Accelerator grouped the code according to its usage and interoperability. It also suggested candidates for migration and ran “what-if” waves against traffic, batch windows, and SLA limits. We selected waves that would provide the greatest reduction of risk during each sprint.
• Checked Design. Every wave had clear rollback points, database diff checks, and contract testing. Feature flags and blue-green delivery made a difference between deployment and release. These features meant that modifications could be dark-launched, watched, and then switched on.

Change went in one way. CI/CD included contract tests, regression packs, and performance probes. If you failed a gate, the wave never moved forward. Success automatically generated new maps and dashboards, facilitating coordination among planning, engineering, and QA teams.

As a result, the client and we were fully satisfied with the course and results of the project. In particular, AI-assisted judgments, instead of relying on manual discovery, reduced the planned modernization time by 70%.

Additionally, the execution of tasks and the assessment after modernization were successful, demonstrating the quality of the decisions made.