AI-native learning platform

Pause the lesson. Ask anything. Watch the answer.

Niyabh Labs lessons are not recorded — they are written by agents and rendered in the browser as they play. So when a learner stops mid-lesson and asks a question, the answer comes back as a new video in the same player, in the same visual language, and the lesson carries on from the exact moment it stopped.

See how it works For creators

First sound of an answer in ~5 seconds · whole answer in ~12

Linux, the shell and you / Unit 2 · Everything is a file

LIVE RENDER

Unit 2 · The shell

Everything is a file

A file is just a stream of bytes with a name
Directories are files that list other files
Your keyboard, disk and screen are files too

Your disk is reachable as a file called /dev/sda.

Lesson paused at 02:14 Back to lesson

/dev/sda Ask about this moment

So why is a hard drive a file too?

Ask

Enter to ask · Esc to go back

Your question

So why is a hard drive a file too?

Writing your answer

Your answer So why is a hard drive a file too? Back to lesson at 02:14

Your program/dev/sdaKernel driverThe disk

Your program only ever talks to a file — the kernel turns that into a disk instruction.

02:14 / 08:30

Ask

Ask the lessonGenerated, not recordedWord-level audio syncInteractive assessmentsDraft → publishRendered at runtimeAsk the lessonGenerated, not recordedWord-level audio syncInteractive assessmentsDraft → publishRendered at runtime

Why we built it

Learning stalls at the
question nobody answers.

Static content has one job — deliver the same thing to everyone. Learning needs the opposite: content that reacts to the person in front of it.

Your question has nowhere to go

A textbook cannot hear you. A recorded video cannot either. The one moment a learner is most engaged — the moment they are confused — is the moment every other format goes quiet.

One pace for everyone

Pre-recorded content is cut once for an imaginary average learner. Too fast for some, painfully slow for others, and impossible to re-cut for the person actually watching.

Watching is not learning

Passive playback proves nothing. Learning needs the lesson to stop, ask something back, and wait for a real answer before it moves on.

The flagship feature

Ask the lesson.

Not a chat window bolted onto a video. The learner asks in the player, about the exact moment they are looking at, and the platform replies in the only format the lesson speaks: a narrated, animated video — generated on the spot.

Pause anywhere

The learner hits Ask. The lesson pauses on the exact frame and the player reads that moment — what is on screen, what was being said, and which part of the diagram was pointed at.

Ask it however you like

Type the question, or tap the microphone and speak it. Live speech recognition handles Chrome and Edge; anywhere else the question is recorded and written out for you.

The answer is made, not looked up

One to three new scenes are written, voiced, assembled into a timeline and validated — a real generated video, in the same player, in the same visual language as the lesson.

The lesson carries on

Replay the answer, ask a follow-up, or go back — and the lesson resumes from the exact millisecond it stopped. Every earlier answer stays on the timeline as a blue marker.

3.5 s

First scene written

~5 s

First sound of the answer

12 s

Whole answer generated

Measured on the interactive-lesson proof of concept

Click a box in a diagram to ask about that specific part

Answers start playing on the first chunk of audio, before generation finishes

Sentence-level captions, aligned to the narration

Follow-up questions chain from the answer you just watched

Draggable timeline, ±10 s, restart, full screen, keyboard shortcuts

Cancel at any point and drop straight back into the lesson

Under the hood

How an answer
gets made.

Five stages, all of them streaming. Nothing waits for the stage after it to finish before the learner sees something happen.

01
Capture the moment
The paused frame is turned into context: the slide on screen, the sentence being spoken, the lesson so far, and the element the learner pointed at.
02
Write the scenes
A reasoning model writes one to three scenes as structured JSON, streaming them out one at a time. It is handed the engine’s own visual catalogue, so it can only choose things the engine can actually draw.
03
Check and repair
Every scene is validated against the engine’s schema and anchor conventions as it arrives, and repaired if it drifts — before a single pixel is drawn.
04
Give it a voice
Each scene is narrated by a generative speech model and streamed back as audio while the next scene is still being written. Playback begins on the first chunk.
05
Assemble the timeline
Exact audio lengths replace the estimated timings, effects are re-anchored to the words they belong to, and the finished topic is stored so replays are frame-accurate.

one answer, on the wire

POST /api/askquestion + the paused moment
scenescene 1 written — estimated timeline
audionarration streams in, playback starts
audio-endexact timings replace the estimate
doneanswer stored, ready to replay

Two players run side by side, because the lesson must not restart. The lesson is paused and kept in memory while the answer plays in its own engine — then handed back at the exact millisecond it stopped.

The rendering engine

Not a video file.
A timeline of commands.

Every lesson and every answer is played by the same framework-agnostic TypeScript engine. It maps a script of timed actions onto an animation loop synced to the narration — which is precisely why new content can be generated into a lesson that is already playing.

A timeline, not a file

Lessons are a sequence of timed commands executed in the browser against a native resolution — there is no mp4 to encode, host or re-cut.

Word-level audio sync

Actions fire relative to specific spoken words, so a highlight lands on the word that explains it rather than a hard-coded second.

Plugin architecture

The core owns time, state and the control surface. Everything visible is rendered by a plugin, so new kinds of content plug in without touching the engine.

Interactive pausing

A plugin can halt the animation loop entirely and hold the lesson until the learner actually does something — the mechanism assessments are built on.

Automatic captions

Captions are extracted from the narration and displayed as whole sentences, in time with the audio. Nothing to author by hand.

Responsive and framework-agnostic

One native resolution scales to any container, and the engine drops into React, Vue, Angular or plain JavaScript — including a headless, chrome-free mode.

Nine plugins do the drawing

the core only owns time and state

Text

Hero statements, title + points, numbered lists, progressive reveal

Visual

Animated explanatory diagrams from a published template catalogue

Code

Syntax-highlighted, walk-through code blocks

Terminal

Typed commands, streaming logs, loading states

IDE simulator

A coding environment acted out on screen

Math

Equation formatting and step-by-step sequences

Data visualisations and generated SVG

GSAP

Multi-element, choreographed DOM animation

Assessment

Quizzes that stop the video until they are answered

Diagrams the writer is allowed to ask for

The visual plugin publishes a machine-readable catalogue of what it can draw. The model that writes lessons and answers is handed that catalogue — so it can never ask for a diagram the engine cannot render.

Flow diagram Layer stack Comparison Hierarchy tree Cycle diagram Annotated box Architecture diagram

How a course is built

An article in.
A lesson out.

Content is authored as writing, because writing is what agents and humans can both reason about. The video the learner watches is generated from it.

Level 1

Course

The whole subject, with its own pricing profile — free, one-time or subscription.

Level 2

Unit

A chapter-sized grouping, brainstormed with an agent before anything is written.

Level 3

Unit item

The smallest learnable atom: an article as the written source, and the generated video the learner actually watches.

Article Topic script Rendered lesson

The written source stays the source of truth. Regenerate the lesson and every learner gets the better version — no re-shoot, no re-edit, no re-upload.

Assessments

The video stops until
you answer.

An assessment is not a quiz page after the lesson. It is an action inside the lesson that halts the animation loop — the lesson genuinely cannot continue until the learner responds, and the submission is recorded against their progress.

Multiple choice

One right answer among distractors written alongside it.

True / false

Fast checks on the claim just made on screen.

Fill in the blanks

Recall the exact term, not just recognise it.

Order the items

Prove the sequence is understood, not memorised.

Match the following

Connect concepts to their definitions or examples.

Generated from the article alongside the lesson · published together · timed submissions tracked per learner

Two surfaces, one platform

One place to author.
One place to learn.

Creators work in the Studio with agents doing the heavy lifting. Learners get the published result. Nothing half-built ever reaches a learner — every piece of content is a draft until it is deliberately published.

For creators

Niyabh Labs Studio

Authoring with agents, not a blank page. Brainstorm the course, shape the units, generate the articles and the assessments — then edit what you got and publish when it is right.

Brainstorm a course and its unit structure in a live agent session
Watch plans and content stream in as they are written
Pick a plan, drop the items you do not want, refine the rest
Generate articles and every assessment type from the article itself
Everything is a draft until you publish — nothing half-built reaches a learner
Upload and manage your own assets

For learners

The learner app

Discover a course, enrol, and learn from lessons that are rendered for you as you watch — with the freedom to stop and ask at any point.

Browse and enrol in published courses, free or paid
Watch engine-rendered lessons with captions and full playback control
Pause and ask a question, and get a generated video answer
Take assessments built into the lesson itself
Progress tracked per video, per unit and per course
Keep a wishlist of what to learn next

Rendering plugins

Diagram templates

Assessment types

~0 s

To the first spoken word of an answer

Questions

The honest answers.

The platform is in active development and the ask-the-lesson flow runs today as a proof of concept. These are the things people ask us first.

Are the lessons pre-recorded videos?

No. A lesson is a script of timed commands — text, diagrams, code, terminals — executed live in the browser by our rendering engine, narrated by generated speech. There is no video file to encode or host, which is exactly why a new scene can be added to a lesson mid-playback.

How long does a learner wait for an answer?

On the proof of concept: the first scene is written in about 3.5 seconds, the first sound plays at about 5 seconds, and the full answer is finished in about 12. The answer starts playing on its first chunk of audio rather than waiting for the whole thing to be generated.

Can the AI invent a visual the engine cannot draw?

No. The model that writes answers is given the engine’s own visual catalogue, so it can only pick templates that exist. Every scene is then validated against the engine’s schema and repaired if needed before it reaches the screen.

Does the lesson lose its place?

Never. The lesson player is paused and kept in memory while the answer plays in a second player. Going back returns you to the exact millisecond you left, and earlier answers stay on the timeline as markers you can replay.

Who writes the courses?

Creators do, with agents doing the heavy lifting in the Studio. They brainstorm the course and its units, generate the articles and assessments, edit the drafts, and publish. Learners only ever see published content.

What subjects can it teach?

Anything the plugins can express: prose and structured text, mathematics, code walkthroughs, terminal sessions, IDE simulations, data visualisations, and seven kinds of explanatory diagram. New subject needs get a new plugin rather than a workaround.

How do assessments fit into a video?

An assessment is a plugin action that stops the animation loop entirely. The lesson does not continue until the learner answers, and the submission is recorded against their progress.

Is it accessible?

Captions are generated from the narration and shown as whole sentences in time with the audio. Playback is keyboard-driven — space to play or pause, arrows to skip, F for full screen — and the whole player respects reduced-motion preferences.

Come and interrupt a lesson.

The fastest way to understand Niyabh Labs is to watch someone stop a lesson mid-sentence, ask something awkward, and get a real answer back. We will walk you through it.

Book a walkthrough Start from the top