I am extending the Huffaz Quran Analyzer with an AI module that can read a user’s handwritten verses in real time. The first goal is to identify whether the writing matches classical Arabic calligraphy—specifically Naskh, Diwani, and Thuluth—and to score the handwriting accuracy against an ideal reference. Once handwriting is verified, the same captured text will be compared letter-for-letter with the original verse so I can gauge the user’s memorisation accuracy from the text itself rather than by timing or repetition counts. What I need from you is an end-to-end solution that plugs straight into the existing Flutter-based front-end: • A recognition engine (Python, TensorFlow, or similar) trained to classify Naskh, Diwani, and Thuluth scripts and output an accuracy score. • A follow-up matcher that aligns the recognised text with the source verse to return a memorisation accuracy percentage. • Clean API endpoints plus brief integration notes so my current mobile team can call your model locally or via REST. The system must cope with common variations in stroke thickness, pen angle, and mobile-camera shadows. I will provide sample pages in each script along with their ground-truth transcripts to bootstrap your training set. Acceptance will be based on: 1. ≥ 90 % correct style classification across the three scripts on a blind test set. 2. ≤ 5 % character-level error when comparing recognised text with the supplied verse. 3. Dockerised build or ready-made model weights with documented dependencies. If you have prior work in Arabic handwriting recognition or calligraphy analysis, I’d love to see a quick demo link or repo.