I’m looking for an experienced Python developer to help improve a text-based listing classification pipeline. The system already works and combines rule-based logic with lightweight ML, but accuracy has plateaued due to logical edge cases rather than model choice. The core challenge is correctly understanding listing intent and product type in noisy, real-world marketplace text (e.g. distinguishing a full vehicle listing from auto parts, electronics, or services, especially in short or ambiguous descriptions). This is NOT a “train a bigger model” task. What I need help with: - Auditing and restructuring the classification pipeline logic - Improving rule ordering and decision flow (product type vs sale intent) - Refactoring brittle keyword rules into more robust, extensible logic - Reducing false positives caused by contextual phrases (e.g. “engine in good condition”) - Improving precision/recall through logic, not heavy ML Current stack: - Python - pandas - regex-based rules - some existing ML components (already implemented) Nice to have: - Experience with text classification or marketplace data - Familiarity with messy, user-generated content - Arabic-language context is a big plus (but not mandatory) Deliverables: - Refined Python classification modules - Clear comments and reasoning behind decisions - A short write-up explaining changes and accuracy improvements I’ll provide the existing repo, data samples, and current benchmarks. Quick feedback and iteration guaranteed.