গণিতমিত্র AI (Ganitmitra AI)

A small, focused Bengali elementary math solver. It reads a natural-language problem (Bangla), predicts the operation with a lightweight ML model, computes the answer, and shows step‑by‑step reasoning in Bengali.

What’s inside (at a glance)

• Web UI (web/): Tailwind CSS + vanilla JS front‑end
• API endpoints:
  • web/solver.php (POST) — solve a problem
  • web/generate.php (GET) — sample problems from dataset.csv
• Core logic (PHP):
  • trainer.php — trains a Rubix ML text classifier
  • solver.php — loads the model, predicts operation, extracts numbers, computes result, builds steps
  • helpers.php — normalization, digit conversion, number extraction
  • result.php — Bengali step rendering

Requirements

• PHP 8.0+
• Composer
• A local server (Laragon/XAMPP) or PHP’s built‑in server

Install

1. Clone and install dependencies

git clone https://github.com/arafat-web/ganitmitra-ai.git
cd ganitmitra-ai
composer install

2. Prepare dataset

• Place your datasets (CSV) in the project root.
• Used by this repo:
  • dataset.csv — used by web/generate.php to show example problems
  • dataset_new.csv — used by trainer.php to train the model

Tip: Keep one canonical dataset if you prefer (adjust scripts accordingly).

Train the ML model

php trainer.php

• Produces bengali_math_model.dat on success (saved when quick check accuracy > 90%).
• Trainer prints a short report (classes found, quick accuracy check).

Run locally

Option A — with Laragon/XAMPP: place the project under your web root and visit /web/

Option B — PHP built‑in server:

cd web
php -S localhost:8000

Open http://localhost:8000 in your browser.

Use the API directly

POST a Bangla problem to the solver endpoint:

curl -X POST \
  -H "Content-Type: application/x-www-form-urlencoded" \
  -d "problem=রাহিমের কাছে ৫টি আম আছে। করিম তাকে ৩টি আম দিল। এখন তার কাছে কতটি আম?" \
  http://localhost:8000/solver.php

Example JSON response (trimmed):

{
  "success": true,
  "problem": "...",
  "bengali_answer": "৮",
  "steps": ["..."],
  "method": "ml_model"
}

How it works (ML)

• Feature extraction: WordCountVectorizer(1000, 2) + TfIdfTransformer
• Classifier: RandomForest (100 estimators)
• Label space (operations): add, subtract, multiply, divide, divide_ceil, multi_step
• Runtime flow (solver.php):
  1. Extract numeric values from text (supports Bengali digits, decimals)
  2. Predict operation class with the trained model
  3. Compute result for that operation
  4. Build human‑friendly Bengali steps

Frontend (web/index.php):

• Clean, mobile‑friendly UI (Tailwind CDN + Font Awesome)
• Enter key triggers solve; Ctrl/Shift+Enter adds a newline
• Shows step‑by‑step solution and a bold Bengali answer
• Loads fresh example problems from generate.php after each solve

Project layout

ganitmitra-ai/
├─ web/
│  ├─ index.php           # UI (Tailwind CDN, Font Awesome), calls web/solver.php
│  ├─ solver.php          # JSON endpoint for solving
│  ├─ generate.php        # Returns random problems from dataset.csv
│  └─ style.css           # Bengali font + small UI polish
├─ trainer.php            # Trains and saves bengali_math_model.dat
├─ solver.php             # ML-based solver (loads model, predicts op, computes)
├─ helpers.php            # Text normalization, number extraction/conversion
├─ result.php             # Builds Bengali step-by-step explanations
├─ dataset.csv            # Example problems for UI suggestions
├─ dataset_new.csv        # Training data for trainer.php
├─ bengali_math_model.dat # Saved model (created after training)
└─ composer.json

Dependencies (Composer)

• rubix/ml (^2.5) — ML pipeline, training, and inference
• league/csv (9.24.1) — CSV utilities (e.g., generate/show dataset)
• mpdf/mpdf (^8.0) — Included for potential PDF export (not used in core flow yet)

Customize and extend (ML)

• Add more labeled examples to dataset_new.csv for operations that underperform
• Tweak vectorizer size or n‑grams in trainer.php
• Try different classifiers (e.g., Naive Bayes, SVC) in trainer.php
• Unify dataset usage (adjust generate.php or trainer.php to the same CSV)

Troubleshooting

• Model file not found

  • Run php trainer.php to generate bengali_math_model.dat.

• Low accuracy

  • Enrich and balance dataset_new.csv; increase vectorizer vocab; try another classifier.

• JSON parse error in UI

  • Check web/solver.php logs; ensure no PHP warnings leak into JSON (buffering+headers are set).

• Mixed dataset files

  • Remember web/generate.php uses dataset.csv while trainer.php uses dataset_new.csv.

• Rubix ML errors you may see

  • "Regressors require continuous labels, categorical given" → use a classifier (you already do) or ensure labels are strings for classification.
  • Class "Rubix\ML\Classifiers\GradientBoostedTrees" not found → install a Rubix version that contains it or use available classifiers (e.g., RandomForest).

Contributing

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

For issues and feature requests, use GitHub Issues.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Connect with me