AI Handwritten Answer Sheet Evaluation & PDF Markup System (Node.js / Vision LLM / OCR)

We are building an automated evaluation engine for handwritten student exam papers. We need an experienced AI/Full-Stack engineer to develop an end-to-end pipeline in Node.js that accepts scanned handwritten PDFs, performs OCR/HTR to map word-level coordinates, evaluates the content using Multimodal Vision LLMs, and programmatically renders high-precision visual annotations, margin callouts, checkmarks, and pointer arrows directly onto the output PDF.

The Result I want is attached Below in the PDF. This is what I want to achieve.