Daniel Bedeleanu 26b38ea27c feat: enhance OCR matching with fuzzy logic and refined AI prompt ...

AI Prompt Improvements:
- Standardized Item name format: [size] type vendor connector part_number
- Clear examples: '5m Patchcord LC-LC', '256GB SSD Samsung SAS', '128GB DDR4 Hynix'
- OCR key format: uppercase, space-separated tokens, includes variations/abbreviations
- Excludes diameter/mm measurements from Item field
- Allows 'HP'/'HPE', 'DDR'/'DDR4', 'NVMe'/'NVME' variations in OCR keys

OCR Matching Logic:
- Implemented Levenshtein distance fuzzy matching (allows 1-2 char differences)
- Priority 0: Exact OCR key match (1000pts) + fuzzy match fallback (800pts)
- Priority 2: Part number exact (200pts) + fuzzy match (150pts)
- Priority 3: Token-based with fuzzy tolerance (50pts exact, 30pts fuzzy)
- Removes whitespace for comparison (e.g., 'LCLC' matches 'LC-LC')

Resolves OCR identification failures from minor text variations.

2026-04-17 12:12:26 +03:00

7.6 KiB

Raw Blame History

View Raw