1. A customer request portal was created
A portal was prepared where the customer could submit bugs, improvements, or feature requests in their own words. This moved requests away from scattered communication channels into a traceable entry point.
An automated development flow from customer request to production release
An automation flow developed for collecting bug reports and feature requests via a customer portal, converting them into technical tasks using LLM, passing them to AI-assisted development automation, running code review checks, validating in a test environment, and deploying after customer approval.
A portal was prepared where the customer could submit bugs, improvements, or feature requests in their own words. This moved requests away from scattered communication channels into a traceable entry point.
Descriptions submitted through the portal were analyzed by an LLM. Customer-written requests were converted into technical work items that could be executed by a developer or automation tool.
The generated technical items were passed into an AI-assisted development automation flow. The system was structured to prepare the required code changes based on the request.
Prepared changes were passed through a separate code review step. The LLM evaluated the changes with an accept/reject logic and reported the decision together with its reasoning.
Completed changes were routed to the test environment. In the used infrastructure, deployment was automated; a similar structure could be adapted to Jira, Jenkins, or comparable CI/CD processes in other projects.
The test environment was checked to confirm whether the application was running correctly. If compile or runtime errors occurred, the error output was sent back into the development automation to trigger a revision loop.
Once the application was running correctly, the relevant development item was presented to the customer for review. This ensured that technical changes were validated by the customer before publishing.
If the customer approved, the publishing process was started. If not, the revision request was fed back into the automation flow and the process continued in a controlled way.