We're building a tool for you to learn anything. To search the bounds of human knowledge & figure things out for you. Agentic learning. Built for the best.
Our Vision
1. Source → Learning Goals → Customizable Curriculum
- AI scans the source material (e.g., this neuroscience paper).
- Generates a "Learning Goals" map: "By the end, you'll understand…" (e.g., vapor self-administration, reinforcement schedules, sex differences in addiction, experimental design).
- Organizes as a navigable curriculum tree/graph (NOT just linear) that users can:
- Expand for details ("What's a fixed ratio?")
- Collapse/hide less relevant branches
- Reorder, bookmark, or request new goals ("Add 'neurobiological basis of reinforcement'")
2. Concept Map, Not Just a Linear Path
- Clickable, web-like map: Every topic/node can be clicked to:
- Zoom in ("Deep Dive: Preclinical animal models")
- Zoom out ("How does this fit into addiction science?")
- Jump to related nodes ("THC vapor in rats" → "THC metabolism in humans")
- Contextual breadcrumbs: Always see "where you are" and "how you got here."
3. Dynamic, On-Demand Deep Dives
- Every node has:
- Core summary
- Links to primary sources or figures
- Option for "Explain Like I'm 5", "Show Details", or "Give Me a Real-World Example"
- Optional "See context"/"See prerequisites": ("Wait, what is 'operant conditioning'? → Click to drill down")
4. Active Practice & Feedback
- Micro-quizzes: After any concept, try a 2-question check ("What does FR5 mean?").
- Socratic prompts: "Why do you think pre-exposure mattered for the rats?"
- Instant feedback: Not just right/wrong, but:
- Detailed correction, hints, or explanation
- Scaffolded follow-up if you get stuck
5. Cross-Linking & Transfer
- Every concept cross-references:
- Other uploaded sources ("SSA in this compiler paper" ↔ "SSA in this superoptimizer paper")
- General foundational knowledge ("SSA" node links to a general "Compiler Concepts" topic)
- Allows transfer: See how a concept works in other contexts or fields.
6. Interface and UX
- Main screen: Interactive concept map (nodes = concepts, edges = dependencies/relationships).
- Click node: Side panel pops up with summary, quiz, "deeper dive," links to paper, and related concepts.
- Navigation bar: "Home," "My Learning Goals," "Bookmarks," "Recent Activity."
- Feedback panel: AI comments on your answers in real time, with suggested resources or next steps.
- User controls: Search bar, progress tracker, export options, history/back button, meta-layer to see your learning journey.
Current State
What we have now: Text → Curriculum. But it doesn't generate the curriculum well - and not entirely in the cloud either! We'll probably need to change that, so that it can run agents well.
We want to build something that takes a paper & analyzes it. The best fit for that is probably Gemini!
Future Functionality
- Create a new space
- Upload files to it → The AI will run over it & you'll get your results
- You can leave the page, even
Minimum Viable Agent Set (for MVP)
If you want to launch a prototype, focus first on:
- Source Analysis Agent
- Curriculum Designer Agent
- Interactive Tutor Agent
- Assessment & Personalization Agent
Add the Deep Dive and Meta-Map agents as you scale.
Example Interaction: All Agents in Action
- User uploads a paper → Source Analysis Agent breaks it down
- Curriculum Designer Agent proposes a concept map/learning path
- Interactive Tutor Agent begins teaching, adapting based on student interaction
- Assessment Agent gives micro-quizzes and tracks understanding
- If user gets stuck, Deep Dive Agent jumps in
- Meta-Map Agent lets the user see their knowledge network and progress