Fix EdTech Parsing Pipeline

I need a senior-level Python engineer who can dive straight into our existing Flask + MySQL codebase and tighten the entire question-conversion pipeline. The first area to resolve is keeping the question prompt, tile bank, interactive worksheet and marking-scheme solution perfectly synchronised. Every phrase must land in the right slot, every ledger row must retain its structure and every derivation step must appear in order.

The only source of truth you should check against is the original CBSE past-paper question together with the official marking scheme. When in doubt, those documents win over anything currently stored in our database.

Your fixes must work across all question types—Accounting tables with ledger and journal entries, Maths and Science calculations, and long-answer theory explanations—so please think broadly about template detection, DOM parsing, NLP-based phrase segmentation and tolerance for mixed layouts.

Key deliverables:
• Refactored parsing modules with clear, modular functions
• Automated Playwright tests that prove one-to-one alignment between source, tiles and solutions
• A validation script that flags any unslotted or mismatched text before questions reach production
• Brief technical notes explaining the approach, critical regex/NLP rules and how to extend them

Our stack already includes BeautifulSoup, lxml, SpaCy and pytest; feel free to introduce lightweight helpers if they keep performance snappy. I’ll provide repo access, sample question sets and failing test cases so you can get started quickly.