AI automation in our own operations

AI automation in our own operations

At a glance

Industry: IT and software development (DYN-IT in-house)

Starting point: knowledge spread across Jira, Confluence, GitHub and Teams; a lot of time went into searching, aligning and copying things by hand

DYN-IT's role: Design, build and operation

Period: ongoing

Result: around 36 production workflows, separate dev and prod environments, approval before every go-live

Technology: n8n (self-hosted), RAG with SAP HANA Cloud and PostgreSQL/pgvector, MCP servers, LLM gateway with zero data retention, Microsoft Teams

Starting point

Our engineers spent a lot of time working with information instead of code: finding solutions from earlier discussions, matching Jira tasks with the documentation in Confluence, copying changes between teams by hand. Every switch between systems cost focus and caused errors at the hand-overs.

What we did

  • Search with sources: AI agents answer questions based on our documentation and code (retrieval-augmented generation) and cite where the answer comes from, so every answer can be checked.

  • Meeting to task: summaries go to Confluence, tasks to Jira and notifications to Teams automatically; voice notes become structured Jira tickets.

  • Quality before the sprint: a review workflow checks tasks against the Definition of Ready and the acceptance criteria.

  • Back office: triage of incoming requests with draft replies, invoice processing, classification of tender documents.

  • Operations: separate dev and prod instances, a mandatory error workflow and documentation per workflow, reviewer approval, and our own 16-rule security check for every workflow.

  • Data protection: personal data is redacted before cloud requests, cloud models run only with zero data retention, and for sensitive data we use local models.

Result

  • Around 36 workflows run in daily production.

  • Fewer repeated questions, answers with a verifiable source, tasks enter the sprint complete.

  • What we built and measured in-house we bring to clients as a pilot, including our decision framework for cloud versus local models.

AI automation in our own operations