M9-018M9: Senior ML & AI System DesignFeature Store, Sharding & CI InfraMedium
⚡Flashcard Drill➔Mastery:
Design a Layout-Aware Document ETL & Parent-Child RAG Indexing Pipeline
🎯 Active Recall Self-Check (Draft First, Verify After)
Login-free practice tracker: recall core answer, audit pitfalls, and log review status.
Step 1 · Draft your concise answer & assumptions
Step 2 · Check Pitfalls & Follow-up Inquiries
📐 Mathematical Definition
⚡ One-Sentence Core Answer
Cure chunk-boundary collapse with layout-aware parsing (Docling), header breadcrumb prefixes, sharp 200-token child index vectors, and rich 1,500-token parent expansions.
📌 Key Takeaways
- •L1 Boundary Collapse: naive character splitting divorces table headers from values and scrambles columns
- •L2 Table Indexing: index concise natural-language table summaries linked to intact Markdown payloads
- •L2 Small-to-Big Index: score crisp 200-token child chunks cleanly, hydrate full 1,500-token parent sections
- •L3 Hierarchy Prefixes: inject H1/H2 title breadcrumbs into every child so pronouns never lose subject context
📐 Mathematical Derivations
Separating index granularity from synthesis granularity optimizes both retrieval SNR and generation coherence simultaneously.
🏭 Production Trade-offs
Persist page-level Docling Markdown into a Silver lakehouse layer so chunking/embedding iterations avoid repeating slow OCR.
⚠️ Common Interview Pitfalls
- ✕Extracting two-column PDFs with layout-blind line readers
- ✕Assuming minor token overlap fixes multi-page table header loss
🎯 Interviewer Follow-up Inquiries
- ?How do you propagate header rows automatically when splitting an 8-page spreadsheet?
- ?How do you measure chunking gains directly via retrieval chunk-boundary recall?
⚡
Test muscle memory on this topic via 3-Layer Flashcards
Active Recall: Mentally formulate Prompt ➔ Math Derivation ➔ Benchmark Implementation in 3 minutes.
📚