A developer using regex to parse quote numbers from PDFs asks whether to switch to machine learning to handle typos and misspellings causing ~1% failure rate. The advice: stick with deterministic logic. Fuzzy string matching (e.g., Levenshtein distance) can handle label misspellings, while quote number typos are harder to solve without additional context. ML is unlikely to improve on an already solid 99% recall, and deterministic approaches are simpler, faster, and more maintainable. The post also clarifies that 100% recall is not a realistic expectation even with deep learning.
1 Impression