Wizard Intelligence Platform

Project Vision

This project is part of a larger goal: building a full-stack, end-to-end AI system inspired by the Harry Potter universe.

The current implementation focuses on spell classification, but the long-term vision is to expand this into a broader platform that integrates:

• machine learning models
• data pipelines
• backend APIs
• interactive frontend experiences

---

Planned Features

Future components of this platform include:

• Sorting Hat Model

  • Classify users into Hogwarts houses based on personality data

• Spell Recommendation System

  • Suggest spells based on user traits or context (content-based filtering)

• Wizarding World Analytics Dashboard

  • Visualize patterns in spells, houses, and character traits

• Character Chat Interface

  • Interact with AI personas (e.g., Hermione, Snape) using NLP

• Audio Processing (Future)

  • Detect speech vs magic-related content (ties into audio ML work)

---

Current Module: Spell Classifier

The current version implements a baseline NLP classifier that predicts whether a spell is:

• dark
• non-dark

This module demonstrates:

• data cleaning and preprocessing
• feature engineering (TF-IDF)
• model training and evaluation
• API deployment with FastAPI
• frontend integration

This serves as the foundation for expanding the system into more advanced ML applications.

How It Works

1. User enters a spell description
2. Frontend sends request to FastAPI backend
3. Backend processes text using trained vectorizer
4. ML model predicts:
   • dark
   • non-dark
5. Result is returned and displayed

---

Example

Input: Avada Kedavra killing curse

Output: dark

---

FastAPI Backend

The project includes a FastAPI backend that serves the spell classification model.

---

Run Locally

1. Activate environment

source venv/bin/activate

2. Start backend

uvicorn backend.main:app --reload

3. Open API docs

http://127.0.0.1:8000/docs

4. Open frontend

frontend/index.html

API Endpoints

• GET /
  Returns a basic API status message.

• GET /health
  Returns API health status.

• POST /predict-spell
  Predicts whether a spell description is dark or non-dark.

Example Request

{
  "description": "causes unbearable pain"
}

Example Response

{
  "description": "causes unbearable pain",
  "prediction": "dark"
}

Model Notes

• Model: Logistic Regression
• Features: TF-IDF vectorization
• Evaluation: Cross-validation (macro F1 ≈ 0.48)

Limitations

• Small dataset (~70 samples)
• Class imbalance (few dark spells)
• Model tends to favor majority class