Problem
In some cases, document chunks may fail to generate vector embeddings (e.g., due to special characters or model errors).
Proposed Solution
Implement a fallback path:
- Skip problematic chunks
- Log skipped entries
- Optionally default to keyword-based retrieval (e.g., BM25 or text match)
Why it matters
Adds robustness for real-world data ingestion and improves model resilience across document formats.
Problem
In some cases, document chunks may fail to generate vector embeddings (e.g., due to special characters or model errors).
Proposed Solution
Implement a fallback path:
Why it matters
Adds robustness for real-world data ingestion and improves model resilience across document formats.