The world (this same grid is sent to every algorithm)
Real streets (live road data from OpenStreetMap)
Click a start on the map, then a destination nearby, then fetch the road network. The same algorithms will race over the real streets.
Pac-Man (the agent learns to play from the screen — no rules given)
Curve = score per training game. Watch it climb as the agent learns to hoover up pellets and dodge ghosts.
Test the model (ask an RL question — compare the base model vs your tuned one)
Base Qwen2.5-1.5B
Load the model, then ask a question.
Your tuned model ✨
Train it first (SFT → DPO) to see the difference.
Reasoning graph (drag from the palette, connect nodes, build a generate→critique→refine loop)
University (a Dean designs a curriculum, then teaches the model each subject to mastery)
How the University works — start here
- Model to teach — a name for your student model. Skills it masters are saved under this name (reusable later).
- Domain — what it should become expert in (e.g. "reinforcement learning"). The Dean (the API teacher) turns this into an ordered curriculum of subjects.
- Mastery target — how good is "good enough" per subject: the teacher-judged win-rate the student must reach on a subject's exam before moving on.
- Subjects — leave at 0 to let the Dean decide how many subjects the domain truly needs. Caps: it runs autonomously and stops at max minutes / budget ($) (and Subjects if you set one). Press Stop anytime.
- Press Enrol & teach. For each subject it: 🧑🏫 writes in-depth Q&A → 🏋️ trains the student → ✍️ runs an exam (you'll see the loop light up per question) → 🎓 records mastery. The Dean re-teaches weak subjects.
- Watch the syllabus bars fill, then use Model & skills (right) to turn any learned skill on/off or retrain it, and Test this model (below) to chat with your graduate.
No API key? It runs a local stand-in (weaker, free) so you can see the whole flow. Set a real key in Teach LLM for genuine depth.
🏫 Campus — step inside each building to watch the Student 🤖 and Teacher 🧑🏫 work
🎛️ Intervene — steer the run live (only while teaching)
💡 Override a grade with the ✓/✗ buttons on the teacher's grade messages below.
💬 Live conversation — everything the agents actually say to each other
The agents' messages — questions, answers, critiques and grades — will stream here during a run.
Test this model
Now testing: — (train or load a model)
Base model (untaught)
—
Your trained model ✨
—