umputun / newscope / 16818615022

push

github

feat(llm): implement batch processing for AI classification with 90% cost reduction

- Add configurable batch processing with size (default: 10) and timeout (default: 5s)
- Remove 500-character content truncation for improved classification accuracy
- Update OpenAI library from v1.40.3 to v1.40.5 and upgrade model to gpt-5
- Implement ProcessBatch method in FeedProcessor for efficient batch handling
- Add comprehensive token usage monitoring and logging for cost tracking
- Fix async test patterns using require.Eventually instead of time.Sleep
- Add setupBasicItemManagerMocks helper to reduce test code duplication
- Update integration tests to handle asynchronous batch processing properly

The batch processing reduces API calls from ~50 individual calls to 2-3 batch calls
for the same number of articles, providing significant cost savings while maintaining
classification quality through full article content analysis.

100 of 124 new or added lines in 2 files covered. (80.65%)

3625 of 4466 relevant lines covered (81.17%)

28.18 hits per line