Problem Statement

The current embedding and LLM integrations for Gemini (internal/integrations/gemini/) do not handle rate limiting (HTTP 429) or transient network errors. When processing high-volume requests (like bulk indexing or frequent triaging), the pipeline fails immediately upon hitting a rate limit.

Proposed Solution

• Implement a retry mechanism with exponential backoff for:
  • internal/integrations/gemini/embedder.go: Embed and EmbedBatch methods.
  • internal/integrations/gemini/llm.go: All generative methods (AnalyzeIssue, RouteIssue, etc.).
• Ensure retries only occur on transient errors (429, 5xx) and NOT on client errors (400, 403, 404).

Context

Identified during self-hosting configuration.