epic(ai): natural language interface

[MR2011]

Description

This issue tracks the implementation of a natural language AI query interface for Heureka. The goal is to allow users to query vulnerabilities, services, and components using natural language without needing to know GraphQL.

• Backend-only: Introduce a new REST endpoint without frontend changes.
• Provider-agnostic LLM abstraction: Internally we'll use the SAP AI Proxy, but should be able to support different models

Question:

Do we want:

• GraphQL generation approach: Use LLM to generate a query, executed against the existing gqlgen executor.
• Template-Based Query Generation: Develop a set of predefined templates for common queries.

Security Considerations

• Query-only: Validate generated GraphQL contains no mutation keyword before execution
• Prompt injection defense: System prompt is separated from user input using proper message roles