Fix OCR & Supabase SQL / Implement Tesseract and LLM

I have a React-based PDF application that relies on a Supabase edge function to perform OCR and then store the results in Postgres. Two pain-points are blocking release: 1. The OCR routine returns incorrect text for both handwritten and printed pages—missing words, adding noise, or scrambling line order. 2. The subsequent SQL logic sometimes fails to persist the extracted data, leaving gaps in the ocr_results table. Your task is to correct the edge function so it extracts clean, readable text from mixed-content PDFs and to rewrite the storage queries so every page’s result lands in the right rows without duplication or loss. Deliverables • Updated edge-function code (TypeScript/Node or preferred runtime) with improved image pre-processing, Tesseract/other engine settings, or alternative libraries, demonstrated on my sample set. • Revised SQL or Supabase RPCs that insert the OCR output and metadata reliably. • Inline comments or a brief README explaining changes and any new env vars. Acceptance criteria • ≥90 % word-level accuracy on the provided 10-document test suite (handwritten + printed). • All records present once, with correct foreign keys, verified through a simple SELECT count(*) script. • GitHub PR passes existing CI pipeline. If you’re comfortable debugging edge functions, tuning OCR engines, and writing clean Postgres SQL inside Supabase, you’ll find the codebase straightforward and well-documented. I’ll give you repo access and sample PDFs as soon as we agree on timelines.

Project ID: 39719238