Cortex-IQ

AI is often a shortcut. Cortex-IQ is a real study partner. Upload your notes, chat with your docs, get smart quizzes, and track your confidence. Master your materials. Don't just bypass them.

Comment 1

Inspiration

Traditional learning platforms treat every student the same, ignoring their unique learning patterns and knowledge gaps. We envisioned an AI agent that builds a cognitive model of each learner, remembering every interaction to create truly personalized study experiences. Our mission was to build a system where the AI grows smarter about you with every question you ask and every quiz you take.

What it does

Cortex-IQ is a cognitive learning platform with an AI agent that continuously learns your knowledge state, study patterns, and comprehension level. The agent analyzes your chat history, quiz performance, and material interactions to generate contextually-aware quizzes targeting your specific gaps and answers questions based on your evolving understanding. It uses voice-enabled tutoring with emotional intelligence, tracks your learning journey with adaptive recommendations, and leverages multi-cloud AI (Gemini, Snowflake API, DigitalOcean Gradient) to maintain persistent context about your learning profile.

How we built it

Frontend: React + TypeScript with real-time state management to track user learning context across sessions. Backend: FastAPI with contextual AI orchestration - Gemini AI maintains conversation context and embeddings, Snowflake Cortex AI performs vector similarity searches to identify knowledge gaps from user history, and DigitalOcean Gradient AI generates adaptive quizzes based on accumulated user performance data. Voice integration combines Web Speech API with 11 Labs emotional TTS, while the agent stores and analyzes user interactions to build a persistent cognitive learning profile.

Challenges we ran into

Building a stateful AI agent that maintains learning context across sessions required designing a sophisticated user profile system that tracks question patterns, concept mastery, and engagement levels. Ensuring the agent generates increasingly personalized quizzes meant implementing vector-based similarity matching between user weaknesses and course materials across three AI platforms. Balancing privacy with personalization required careful data architecture where the agent learns deeply about users while keeping their learning data secure and portable.

Accomplishments that we're proud of

Created an AI agent that genuinely "remembers" each student - tracking what they've mastered, what confuses them, and how they learn best to generate perfectly-targeted quizzes. Built a cognitive feedback loop where every interaction (chat, quiz, voice question) refines the agent's understanding of the user's knowledge state. Achieved seamless context transfer across three enterprise AI platforms so the agent leverages Gemini's conversational memory, Snowflake's analytical pattern recognition, and DigitalOcean's adaptive generation as one unified intelligence.

What we learned

Context-aware AI requires more than chat history - it needs structured knowledge graphs that map user understanding across topics, time, and confidence levels. Effective personalization demands balancing what the agent knows about past performance with adaptive scaffolding that challenges users at the right difficulty level. Multi-cloud AI architectures become exponentially more powerful when each platform contributes to a shared cognitive model rather than operating independently.

What's next for Cortex-IQ

Implement deep learning models that predict optimal study timing based on individual forgetting curves, with the agent proactively suggesting review sessions before knowledge decay. Add peer learning insights where the agent identifies similar learning patterns across anonymized users to recommend study strategies that worked for others with comparable profiles. Expand the cognitive model to recognize learning styles (visual, auditory, kinesthetic) and automatically adapt content presentation, quiz formats, and explanation depth to match each user's optimal learning mode.

Built With

Share this project:

Updates