Tickets

Context Grabber

When a new task comes in, an analyst most often has to rebuild the system picture from scratch. They open Confluence, draw.io, READMEs and Jira and spend time on what a colleague already figured out yesterday. An agent given the same task collects context itself, but fills the gaps with guesses.

We built a context grabber that generates documentation from code and checks it for distortions and hallucinations. For every service an agent writes contracts, end-to-end scenarios, architecture decisions and a glossary. Everything is stored in Git as plain text files, with a local map of connections between 60 services and 22 external systems on top. The same files are read by teammates and by an agent in the IDE.

In this talk, I will show how it works and give a step by step plan to set it up in your team. I will also cover where generation went wrong and why we did not notice it right away.

Comments ({{Comments.length}})
  • {{comment.AuthorFullName}}
    {{comment.AuthorInfo}}
    {{ comment.DateCreated | date: 'dd.MM.yyyy' }}

    1
To leave a feedback you need to

or
Chat with us, we are online!