Cognito

Your brain's AI architect. Turn raw data into interactive knowledge graphs, dynamic roadmaps, and automated active recall challenges for rapid mastery

Comment

Conventional note-taking tools are often just passive digital "junk drawers." I realized that students and researchers don't just need a place to store data; they need a partner to synthesize it. Our inspiration was to bridge the gap between static information and active mastery by building an AI-powered ecosystem that visualizes connections, builds path-to-mastery roadmaps, and automates active recall.

Cognito is a professional-grade research and study platform that turns raw inputs into a structured second brain.

Knowledge Graph: Every note is a node. Cognito automatically maps relationships between subjects, allowing you to visualize your intellectual growth in a high-performance interactive graph. Intelligence Vault (PDF Ingest): High-fidelity text extraction from complex PDFs with real-time AI refinement. It cleans up technical formatting and structures notes with headers and LaTeX math. Roadmap Architect: My AI builds custom, step-by-step learning roadmaps for any topic. Users can track their progress through a "protocol-style" interface with interactive nodes. Mathematics Workbench: A specialized lab for engineers and scientists supporting full LaTeX and interactive graphing for active visualization of formulas. Logic Lab (AI Exercises): Automatic generation of active recall challenges (Quizzes, Math problems, and Conceptual Deep-dives) based on your own notes.

I architected Cognito for speed and a premium, high-impact aesthetic:

Frontend: Built with Next.js. I used Tailwind CSS for desig. Framer Motion powers our micro-interactions, and @xyflow/react (React Flow) handles the complex graph and roadmap visualization. Editor: A highly customized Tiptap instance with custom extensions for math (MathJax), task tracking, and specialized AI-selection refinement. Backend: A high-performance FastAPI (Python) server handling data persistence with PostgreSQL and SQLAlchemy. AI Engine: We integrated Llama-3-8B to power our real-time streaming services for roadmaps, math solutions, and text refinement. Storage: Direct AWS S3 integration for knowledge assets and image management with a custom-built cleaning utility to ensure storage hygiene.

One of the biggest hurdles was maintaining zero-lag performance while handling large-scale knowledge graphs and real-time AI streaming. We also had to solve the "Context Window" problem: when users have massive notes, sending them to the AI often crashed standard models. We solved this by implementing a custom character-based safety layer and intelligent truncation logic on the backend.

Zero-Placeholders: Every part of Cognito is functional - from the AI-driven roadmap generator to the interactive math graphing tools. Advanced Data Ingestion: Building the PDF Vault with dynamic character-count limits and selection refinement was a technically complex task that we executed successfully. We learned the importance of "Active UX" - ensuring that the interface gives constant visual feedback (like our selection counters and AI thinking indicators). I also deepened our knowledge of visualizing complex data structures and managing real-time websocket/streaming responses in a production-ready Next.js environment. Collaborative Graphing: Enabling real-time "knowledge-sharing" where teams can merge their study graphs into a singular project library. Edge AI: Transitioning some of the lighter intelligence tasks to local models to reduce latency and enhance privacy.

Built With

Share this project:

Updates