Grammar & Spell Checker

Rule-based spell checkers fix typos but cannot see that a sentence is ungrammatical. Language models fix grammar but are heavy and fussy to install — and the usual outcome is a project that works on the machine it was written on and nowhere else.

This project pairs both approaches, then removes the environment problem entirely by shipping the whole thing as an image.

1. Two layers, because they fail differently

   TextBlob handles spelling, where a dictionary is exactly the right tool. A Hugging Face transformer running on Torch handles grammar, where context is required. Each one covers the other's blind spot, instead of asking a single component to do everything.

2. Docker as the actual feature

   A Dockerfile pins Python 3.10 and the full model environment, so docker run is the entire install procedure. For an NLP app with Torch in the dependency tree, reproducibility is not a nicety — it is the difference between a usable project and a README full of caveats.

3. A swappable model

   The correction model sits behind Model.py rather than being inlined into the Flask routes, so it can be exchanged or fine-tuned for domain-specific grammar rules without touching the web layer.

4. Honest about production

   Debug mode is enabled for development only, and the developer notes call for Gunicorn or uWSGI in front of the app before it faces real traffic — documented rather than silently shipped as-is.

1. Build the image

   docker build -t grammar-spell-checker .

2. Run the container

   docker run -p 5000:5000 grammar-spell-checker

3. Open it

   http://localhost:5000

4. Try the sample

   A single sentence carrying four separate errors — verb agreement, a double negative, a countability mistake and tense.

   Input:
     He dont has no idea how many informations
     was missing from the report.
   
   Output:
     He doesn't have any idea how much information
     was missing from the report.