Skip to content

Core flow redesign #8

@avikalpg

Description

@avikalpg

The current flow overuses AI features that lead to unnecessary hallucinations and doesn't show useful enough information.
Here is a list of changes that should be made:

  • Instead of using AI to navigate all the dependencies, we should use tree-sitter or LSPs for that, as long as it is a supported programming language (we can use AI as a fallback)
  • List all the possible ways in which different services talk to each other (REST API calls, gRPC/tRPC, socket connections, DB-writes to shared memory, etc.) and get the rule-based engine to identify all such instances. This should trigger an AI agent flow to identify, first, whether this is an internal endpoint or third party; and if it is internal, then AI, with the help of grep-tools, identify those inter-service connections.
  • The connections should be identified for all the affected code blocks in the diff. And the whole code block, with both versions, should be sent to the LLM for it to be able to identify what has changed. And this should be stored along with the diff specific to that code block in an object that can then be shown to the user.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions