AI Sentence-Level Plagiarism Tool

The project is to build a web-based plagiarism checker that can ingest large documents (10,000+ words) in PDF or DOCX format, scan every sentence, highlight any matches it finds, and instantly generate an English-only rephrased version that preserves the author’s intent. The user should be able to upload three to five files in a single batch, see colour-coded plagiarism highlights on screen, accept or reject each auto-rewrite, then download the fully updated document as a single DOCX file.

Core expectations
• Accurate, sentence-level comparison against the major public web and academic indexes.
• Fast in-memory processing for files of 10k words or more without timeouts.
• High-quality rephrasing powered by modern NLP (e.g., Transformer or GPT-style models) with meaning retention checks.
• A clean browser interface: drag-and-drop upload area, progress indicators, side-by-side original vs. rewritten sentences, final one-click DOCX export.

Deliverables
1. Fully functional application (frontend + backend) deployable on a standard cloud VM or Docker.
2. Source code with installation guide and model-training or API-integration notes.
3. Test suite covering PDF/DOCX ingestion, detection accuracy, rephrase quality, and batch export.
4. Brief user manual outlining the workflow from upload to download.

Acceptance criteria:
• For a provided test set, the tool must flag ≥95 % of intentionally plagiarised sentences.
• Each auto-rewrite must pass a follow-up scan with ≤5 % residual similarity to the same public sources.
• Batch processing of five 10k-word DOCX files must complete in under five minutes on a 4-core machine.

Feel free to suggest the tech stack—Python + FastAPI, Node + Express, or another modern solution—as long as it meets these functional benchmarks and remains easy to maintain.