English Spelling (Typo) Correction System — BART Seq2Seq
Personal Project · FPT University
Built an AI text corrector that automatically identifies and fixes character-level spelling errors in English sentences.
Fine-tuned facebook/BART-base on a synthetic dataset of 183,653 noisy text pairs generated from a 30-book corpus, reaching a 75.4% F1-score.
Deployed the model as a Gradio web app.
- PyTorch
- BART
- NLP
- Gradio
- Hugging Face