Semantic Quiz - AI-Powered Answer Grading

📖 How Does Semantic Quiz Grading Work?

🔤 Text to Numbers

The answer is converted into a 384-dimensional vector (a list of 384 numbers) using the all-MiniLM-L6-v2 transformer model. This captures the meaning of the text, not just the exact words.

📐 Cosine Similarity

We compare the answer vector to the expected answer vector by measuring the angle between them. Smaller angle = higher similarity. This means paraphrased answers still get credit!

similarity = (A · B) / (||A|| × ||B||)

Dot product divided by the product of magnitudes

🧠 Why Embeddings?

Unlike exact text matching, embeddings understand that "Paris is the capital" and "The capital of France is Paris" mean the same thing. This is the power of semantic understanding!

⚡ 100% Browser-Based

Everything runs locally in your browser using WebAssembly (WASM). No data is sent to any server - your answers stay private. The model is cached after first load for instant startup.

🎯 Scoring System

Your score is determined by how semantically similar your answer is to the expected answer:

Similarity	Rating	Points	Meaning
≥ 85%	Excellent!	1.0	Your answer captures the core concept accurately
≥ 70%	Good!	0.8	Good understanding with minor gaps
≥ 55%	Partial	0.5	Some relevant concepts mentioned
< 55%	Try Again	0	Answer doesn't match the expected concept

🧠 Semantic Quiz