🎓 ScholarPath — AI-Powered Document Verification System

An intelligent document verification platform for scholarship systems using OCR + Machine Learning + Rule Engine + RAG (Retrieval-Augmented Generation).

---

🚀 Overview

ScholarPath automates the verification of student documents such as marksheets and income certificates by combining:

• 🧾 OCR for text extraction
• 🧠 ML model for fraud detection
• ⚖️ Rule-based validation
• 📚 RAG for explainable decisions

👉 The system not only predicts validity but also explains why a document is accepted or rejected.

---

🏗️ Architecture

Upload Document
   ↓
OCR (Text Extraction)
   ↓
Document Type Detection
   ↓
Feature Extraction
   ↓
Rule Engine
   ↓
ML Model Prediction
   ↓
RAG (Rule Retrieval + Explanation)
   ↓
Final Decision

---

✨ Features

• 📄 Supports image-based documents (PNG, JPG)
• 🔍 OCR using Tesseract
• 🧠 ML-based fraud detection (Random Forest)
• ⚖️ Rule-based validation (income thresholds, document completeness)
• 📚 RAG-based explanation system
• ⚡ FastAPI backend with interactive Swagger UI
• 🧩 Modular and extensible architecture

---

🛠️ Tech Stack

• Backend: FastAPI
• ML: scikit-learn (Random Forest)
• OCR: pytesseract + OpenCV
• RAG: JSON-based rule retrieval (extendable to FAISS)
• Frontend: React + Vite (in progress)

---

📂 Project Structure

backend/
├── data/
│   ├── raw/                # Training images
│   ├── labels.json        # ML labels
│   └── rules.json         # RAG rules
│
├── services/
│   ├── ocr.py             # OCR extraction
│   ├── feature.py         # Feature engineering
│   ├── model.py           # ML model
│   ├── rules.py           # Rule engine
│   ├── rag.py             # RAG logic
│   └── llm.py             # (future use)
│
├── main.py                # FastAPI app
├── train.py               # Model training script

---

⚙️ Installation

git clone https://github.com/your-username/scholarpath.git
cd scholarpath/backend

python -m venv venv
source venv/bin/activate   # mac/linux
venv\Scripts\activate      # windows

pip install -r requirement.txt

---

▶️ Run the Application

uvicorn main:app --reload

Open:

👉 http://127.0.0.1:8000/docs

---

🧪 API Usage

POST /verify

Upload an image file.

Example Response:

{
  "doc_type": "marksheet",
  "features": {
    "income": 42,
    "text_length": 322
  },
  "rule_result": {
    "status": "Accepted",
    "reason": "No rule violations"
  },
  "ml_prediction": {
    "is_valid": true,
    "confidence": 0.75
  },
  "explanation": "Accepted. Rule satisfied: Document must contain valid marks and student details"
}

---

🧠 Key Concepts

🔹 Rule Engine

Applies deterministic checks:

• Income thresholds
• Document completeness

🔹 Machine Learning

Learns patterns from labeled data:

• Fraud detection
• Anomaly scoring

🔹 RAG (Retrieval-Augmented Generation)

• Retrieves relevant rules from knowledge base
• Generates human-readable explanations

---

📊 Model Training

python train.py

• Uses features extracted from documents
• Trains Random Forest classifier
• Saves model as fraud_model.pkl

---

⚠️ Limitations

• Small dataset → limited model accuracy
• OCR noise affects feature extraction
• Basic RAG (keyword-based, not semantic yet)

---

🚀 Future Improvements

• 🔥 FAISS-based semantic RAG
• 📄 PDF support
• 🎯 Better feature extraction (name, income parsing)
• 🌐 Full frontend dashboard
• ☁️ Deployment (AWS / Render / Docker)

---

💡 Why This Project Stands Out

• Combines ML + Rules + RAG (real-world architecture)
• Focus on explainability, not just prediction
• Modular design for multiple document types
• Solves a real problem in scholarship verification

---

👩‍💻 Author

Shreya Yadav Machine Learning Engineer

---

⭐ If you like this project

Give it a star ⭐ and share feedback!