Twinbook Creator Studio

Take your ideas further.

Go beyond your videos. Turn them into beautiful visual notes, adaptive quizzes and a memory system that helps your audience remember, in your own space, under your own name.

Reach more learners, see where your content can improve, and share in a royalty pool. Or email us →

Already have a space? Sign in

.twinbook.ai

Choose a style

13 styles · use the arrows to see more

Field notebook

Plants, specimens, nature

Whiteboard

Processes and flows

Paper cutout

Cycles and systems

Scratch paper

Maths and proofs

Textbook

Parts and anatomy

Cutaway

What's inside

Exploded parts

Layers and assemblies

Three panels

Change over time

Before and after

Cause and effect

Magnified inset

Big and tiny scales

Visual analogy

Abstract ideas

Scrapbook

History and events

Chalkboard

Classic explanations

Pick the look of your visuals. Every figure is drawn for your topic and stays editable.

What your video becomes

Your video, as notes with figures.

Even a hard topic like reinforcement learning turns into notes people can follow: the maths written out, and a figure for every key idea in the style you picked.

Scroll the notes to see more →

▶18:24

From your YouTube video

Reinforcement learning from scratch

Your channel · 5 figures

Reinforcement learning

How a machine learns by trying things and keeping score.

1. The agent loop

  • An agent sees a state st, picks an action at, and gets a reward rt+1.
  • Nobody tells it the right move. It only learns which actions pay off.
Figure 1 · The loop between agent and environment (whiteboard)
Figure 1 · The loop between agent and environment (whiteboard)

2. Rewards over time

  • The agent maximises the return: every future reward, with later ones counting less.
  • The discount γ (between 0 and 1) sets how far ahead it plans.
Gt = rt+1 + γ rt+2 + γ2 rt+3 + … = Σk γk rt+k+1
Figure 2 · Why later rewards shrink (scratch paper)
Figure 2 · Why later rewards shrink (scratch paper)

3. Values and policies

  • The value V(s) is the return you can expect from a state. The Bellman equation defines it one step at a time:
V(s) = maxa [ R(s, a) + γ Σs′ P(s′ | s, a) V(s′) ]
  • Repeat that update everywhere (value iteration) and values spread out from the goal. The policy just follows the arrows uphill.
Figure 3 · Values grow near the goal (textbook)
Figure 3 · Values grow near the goal (textbook)

4. Explore or exploit

  • Always picking the best-known action can miss something better. ε-greedy explores with a small probability ε.
  • Learning from each step with Q-learning:
Q(s, a) ← Q(s, a) + α [ r + γ maxa′ Q(s′, a′) − Q(s, a) ]
Figure 4 · Your favourite place or a new one? (visual analogy)
Figure 4 · Your favourite place or a new one? (visual analogy)

5. RL for language models (RLVR)

  • Reinforcement learning with verifiable rewards trains a language model on questions whose answers can be checked, like maths problems or code with tests.
  • A simple checker gives reward 1 for a correct final answer and 0 otherwise. No human judge or learned reward model is needed.
  • The term comes from the Tülu 3 paper (Lambert et al., 2024). DeepSeek-R1 trained its reasoning with rule-based rewards like these and GRPO, which scores each answer against the others in its group:
Ai = ( ri − mean(r1, …, rG) ) / std(r1, …, rG)
Figure 5 · A checker turns right answers into reward (whiteboard)
Figure 5 · A checker turns right answers into reward (whiteboard)

End of notes · back to the video at 18:24

Creator Studio · What your content becomes

Not an AI summary. Your work, kept the way you made it.

Bring your documents and handwritten notes. Twinbook keeps your style, your figures, your tables and your way of explaining, where a generic AI summary would flatten them.

This is what your audience reads, asks about and gets quizzed on.

✦ Powered by our own AI models

Built for real-world material: low-resolution scans, poor lighting, handwritten notes and figures, tables and dense text. The pipeline is tuned for accuracy while keeping it affordable.

(Contact us for a technical report on performance.)

Customize your high-yield note.

One click on presets and the same source becomes whatever you need — in-depth for a first pass, high-level only before an exam, SOAP note for the clinic. Or describe your own style.

Creator Studio · Publish and measure

Turn your content into measurable learning.

Create cohorts and publish videos, readings, and practice as one guided learning experience. Twinbook helps you deliver the course, while Learning Intelligence shows what learners understand, where they struggle, and what you should improve.

See what learners understood, not just what they opened.

Run your own cohort

Invite your community and keep talking where they already are.

DiscordSlackZoomSupported

Creator Studio · Your space

Your brand. Your audience. Your growth.

Give every learning experience a branded page with its own shareable Twinbook URL. See who visits, who signs up, and how your audience becomes learners—all from Creator Studio.

Security and privacy by design

A branded Twinbook page with its shareable URL, next to the audience analytics dashboard tracking visits, signups, and conversions

Ready to publish your first Twinbook?

Tell us what you teach. We'll set up your space with you.

or email us →