Introduction: Academic Oracle—An Adaptive AI Learning Platform Based on Structured Reasoning

Academic Oracle is an AI education platform focused on deep learning, with the core philosophy of "helping students think rather than just providing answers". Through structured reasoning processes, progressive prompting, and a dual-mode (learning + exam) design, it achieves a complete closed loop from deep understanding to exam performance, distinguishing itself from traditional fast-response AI Q&A tools.

Project Background: Evolution from Q&A Tool to Complete Learning System

Academic Oracle started as a chatbot and has now evolved into a complete system integrating "learning mode" and "exam mode". Its design goal is not only to cultivate deep understanding but also to train students' performance under real exam conditions, forming a seamless connection between learning and assessment.

Learning Mode: Structured Learning Process Based on Cognitive Science

The learning mode follows a scientific cognitive process: Question → Think → Prompt → Attempt → Feedback → Pattern Discovery → Insight → Mastery, inspired by the principles of active recall and spaced repetition. Core mechanisms include:

• Active recall priority: Let students recall independently first before showing answers to promote long-term memory;
• Progressive prompting: Layered guidance instead of direct answers to maintain cognitive engagement;
• Feynman Technique: Guide knowledge reconstruction from first principles to ensure genuine understanding;
• Error correction loop: Help students learn from mistakes through pattern extraction.

Exam Mode: Real-Scenario Simulation and Intelligent Assessment System

The exam mode focuses on real exam performance training, with core features including:

• Strict time limits: Simulate exam pressure and train time management;
• Graded assistance control: 0-3 levels of assistance (from no prompts to complete solutions);
• Alignment with scoring standards: Step-by-step scoring logic, providing real-time feedback on actual exam scores;
• Intelligent paper processing: Support for extracting questions and scoring schemes from multiple formats;
• examMemory system: Track weak topics and error patterns, generate structured review lists.

Technical Architecture: Multi-Model Orchestration and Security Assurance

The technical architecture adopts a Gemini-first strategy with OpenRouter as a backup, providing five execution modes (Standard/Fast/Balanced/Agent/Web Search). The upgraded competition logic implements "first valid response wins", and dynamic routing optimizes performance. Real-time knowledge retrieval integrates Tavily (primary) and JigsawStack (backup), activated only on demand. For security, API calls are processed through the Supabase backend, core prompt logic is centralized, sensitive data is encrypted, and a jailbreak detection mechanism is included.

User Experience: Minimal Disturbance Design and Learning Progress Visualization

The interface follows the "minimal disturbance principle", providing context-aware buttons, intelligent text selection suggestions, a unified chat + quiz UI, dark/light modes, and responsive design. The learning dashboard displays user profile, academic level, current topic, and learning grade, including a learning efficiency ring indicator and expandable topic panels (key notes, formula prompts, quiz records, etc.).

Educational Value and Outlook: A New Benchmark for AI-Assisted Learning

Academic Oracle represents the evolutionary direction of AI education tools: shifting from speed to effectiveness, which aligns with contemporary educational research (depth over speed, active over passive, feedback over answers, continuous over isolated). It provides a complete training ecosystem for students preparing for exams like IGCSE/A-Level/AP/SAT. The conclusion points out that it demonstrates the correct way to enhance learning with AI, setting a new benchmark for AI-assisted learning, and its design centered on learning effectiveness will become a reference for educational technology.