README.md

metadata

title: WriteRight NLP Backend
sdk: docker
app_port: 7860

WriteRight NLP API 🚀

WriteRight is a production-grade NLP auto-correction and text refinement backend. It utilizes a multi-stage pipeline combining algorithmic rules with state-of-the-art Transformer models to provide instant spelling, grammar, and stylistic corrections.

🧠 Architecture & Models

The backend utilizes a decoupled architecture to ensure high precision:

1. Tokenization & NER: spaCy (en_core_web_sm)
2. Spell Correction: SymSpell (O(1) lookup, edit distance 1-2)
3. Homophone Resolution: Context-aware POS heuristics
4. Grammar Correction: vennify/t5-base-grammar-correction (T5 Encoder-Decoder)
5. Stylistic Refinement: eugenesiow/bart-paraphrase (BART Sequence-to-Sequence)

By separating Grammar Correction (T5) from Stylistic Refinement (BART), the system prevents prompt-leakage and maintains semantic integrity during paraphrasing.

🔌 API Endpoints

The API is built with FastAPI and is fully documented. Once running, visit /docs for the interactive Swagger UI.

1. Correct Text (POST /api/correct)

Executes the primary multi-stage correction pipeline.

Request: ```json { "text": "Thier going to the store to much.", "run_spell": true, "run_grammar": true, "run_homophones": true } ```

Response: ```json { "original": "Thier going to the store to much.", "corrected": "They're going to the store too much.", "spell_fixed": 0, "grammar_fixed": 1, "homophone_fixed": 2, "confidence": 0.89, "diffs": [...], "processing_ms": 345.2 } ```

2. Refine Text (POST /api/refine)

Paraphrases and refines text for flow and style using the BART model.

Request: ```json { "text": "They're going to the store too much.", "style": "professional" } ```

🛠️ Local Setup & Development

Prerequisites: Python 3.9+

1. Install dependencies: ```bash pip install -r requirements.txt python -m spacy download en_core_web_sm ```

2. Run the server: ```bash uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload ```

☁️ Hugging Face Deployment

This repository is configured for automatic deployment on Hugging Face Spaces via Docker.

• Port Mapping: The Dockerfile routes Uvicorn to 7860 as required by HF Spaces.
• Cold Starts: Upon the very first container initialization, the system will download the T5 and BART model weights (~1.5GB) from the Hugging Face Hub. Initial boot may take 2-3 minutes. Subsequent requests will be processed in-memory.

---