Description:
When a failure occurs during the LLM call in the chat processing pipeline—particularly during the profile extraction phase—the corresponding chat data is lost from both the general_blobs and buffer_zones table.
Expected Behavior:
Even if the LLM call fails, the chat data inserted should persist in general_blobs and buffer_zones for debugging, retrying, or audit purposes.
Steps to Reproduce:
- Trigger chat processing with profile extraction.
- Cause a failure during the LLM call (e.g., by simulating a timeout or API error).
- Observe that the related chat entries are missing from general_blobs and buffer_zones.