Handwritten Answer Sheet Evaluation System

I am looking to develop a comprehensive system that evaluates handwritten answer sheets by comparing them with a model answer. The system should include both mobile and desktop user interfaces, along with a robust backend AI engine. The key requirements are:

1. Mobile App (React Native / Flutter):
- File upload functionality for handwritten answer scripts and model answers.
- View results, including annotated PDFs, marks, and feedback.
- Teacher dashboard for rubric configuration and custom feedback.

2. Desktop Interface:
- Similar functionalities as the mobile app, optimized for desktop use.

3. Backend API (FastAPI / Node + Express):
- Orchestrates the pipeline and stores job status/results.

4. Storage (S3 / GCS):
- Secure storage for PDFs, images, and annotations.

5. Preprocessing Service:
- Converts PDF to images (per page).
- Performs image deskewing, cropping, and page segmentation.

6. Handwritten Text Recognition (HTR) / OCR:
- Converts handwriting into plain text for further processing.

7. Document Understanding & Mapping:
- Maps extracted text to specific questions or parts of the document.

8. Grading Engine:
- Uses rule-based methods and ML/LLM scoring to evaluate answers against a configurable rubric and embeddings.

9. Feedback Generator:
- Generates customized feedback using templates and LLM-based phrasing techniques.

10. Human-in-the-Loop Review UI:
- Allows teachers to review, correct, and retrain the system as needed.

11. Monitoring/Logging System:
- Tracks metrics and logs for monitoring performance.

Workflow Overview:
- Users upload a PDF of a handwritten answer script along with the question or select a question from the database.
- The backend processes the handwriting into text using OCR/HTR.
- The extracted text is mapped to specific questions.
- The grading engine evaluates the answers against a rubric.
- Marks, annotated PDFs, and customized feedback are generated and returned to the user.