Chapter 11 Filing Extraction Pipeline
🏦 AI/ML Engineering Internship at Bondoro Insights
A document AI pipeline that converts scanned, hundreds-of-page Chapter 11 bankruptcy filings into highly accurate structured text feeding LLM pipelines that generate case summaries and pitchbooks for large-cap bankruptcies. Because downstream summarizers faithfully report whatever the extraction says, the system is engineered around a single principle: a confidently wrong value (a silently corrupted figure or a number drifting away from its label) is worse than a loud failure. It scored 99.3% composite accuracy across a 684-document benchmark.
The architecture is a deterministic OCR spine with routed VLM overlays — page-level quality scoring identifies problematic content and routes only those pages to heavier models, so expensive stages run only where they provably win. Computer vision stages recover checkbox states from degraded scans (faded borders, hand-drawn ticks, bleed-through) and de-weld superscript footnote markers that OCR engines misread as digits in financial figures — a class of silent corruption where a single welded character can turn a correct dollar amount into a plausible 10× error. An independent OCR cross-check catches silent line-detection misses from the primary engine. Engine selection was settled through structure-aware bake-offs across 5+ OCR/VLM candidates, and a CI-gating evaluation harness enforces exact match on financial fields and label–value association against hand-verified ground truth. The pipeline runs deterministically and local-by-default on consumer GPUs, with serverless GPU offload for heavy stages and reproducible parity across dev and production environments. (Proprietary codebase — no public repo.)