Visit the live Web App : examcram.azurewebsites.net

Agent-Exam-Cram-The-Active-Recall-Tutor

An interactive multi-agent tutor that uses active recall, fact-verification tools, and automated grading to help you master any subject.

Agent Exam Cram: The Active Recall Tutor

Track: Agents for Good (Education)

Problem Statement

Students, especially undergraduates often struggle with "passive review" reading notes without retaining information. Effective studying requires "active recall," but finding a study partner to quiz you at midnight during exam season is impossible. Manual flashcards are static and cannot explain why an answer is wrong.

Solution Statement

"Agent Exam Cram" is an intelligent study companion available 24/7. Unlike a standard chatbot, it is a Multi-Agent System that acts as both a Tutor and an Evaluator. It uses tool-calling to reference "ground truth" definitions (simulating a textbook lookup) to ensure it grades student answers based on facts, not hallucinations.

Architecture

This project utilizes a Multi-Agent Architecture powered by Google Gemini 2.5 Flash.

• Agent 1: The Tutor (Active)

  • Role: Conducts the interactive quiz loop.
  • Tools:
    • lookup_textbook: Verifies answers against ground truth (Hallucination prevention).
    • give_hint: Provides scaffolding when a student is stuck.
  • Memory: Maintains active session context to track the conversation flow.

• Agent 2: The Evaluator (Passive)

  • Role: Analyzes the full conversation transcript after the session ends.
  • Task: Generates a structured "Report Card" with a letter grade and study recommendations.
  • Concept: Implements the "Agent-as-a-Judge" pattern to evaluate student performance.

Value Statement

This agent reduces the cognitive load of planning a study session. By automating the "quizzing" process and providing a final summary grade, it allows students to focus entirely on recall and understanding.