The distance between an idea in the head of an end user and an actual development effort is usually too large. Meetings, requirement specifications and quote rounds eat time before anything has been built. We have built a chatbot with indirect access to the codebase, it lets the requester describe what they want and get a realistic time estimate immediately, grounded in the actual code.

The problem we solve

Traditionally, requesting a new feature is a slow multi-step process:

• The requester puts the idea into writing or into a meeting
• The development team has to interpret and break it down
• Estimates are produced manually, often with wide uncertainty
• The decision to start development drags out

The result is high friction, long lead times, and estimates built on assumptions rather than facts.

How the knowledge base is built

The AI continuously reads the codebase and maintains an up-to-date knowledge base of markdown files describing:

• Which modules and services exist and how they connect
• The business logic and rules that drive behaviour
• Workflows from the user's perspective
• Data models, integrations and dependencies

The knowledge base updates when the code updates, it does not lag behind, does not go stale, and does not need manual upkeep.

How the chat works for the end user

The requester opens a chat and describes what they want, in the same way you would talk to a colleague:

• Ask how an existing feature works
• Describe a new feature or an adjustment
• Get follow-up questions when scope needs to be sharpened
• Attach screenshots, examples or links

The chat answers system-specific questions from the knowledge base, not from general knowledge.

Real-time estimates grounded in real code

Once the change is described, the AI matches it against the codebase to estimate the effort:

• Which files and modules need to change
• Whether APIs, data models or integrations are affected
• Need for tests, migrations and documentation
• Risks and dependencies on other development

The estimate is presented as a range and updates immediately if the requester changes scope.

From idea to development

When the estimate feels right, the requester can approve it directly in the chat:

• The request goes into the development queue with full context
• The developer gets a structured spec, not a loose idea
• The requester can follow status in the same chat
• Deviations from the estimate are reported back transparently