We use cookies and similar technologies

Coursfy uses essential cookies and browser storage to run the site. The optional categories stay off until you switch them on: analytics turns on error monitoring, and marketing and cross-device analytics load nothing today. You can accept all, reject non-essential, or manage each category. AI and voice processing are set in Privacy settings. Privacy Policy · Cookie Policy

Skip to main content
Coursfy
HomeAcademiesCoursesPricingLuminaCreate an academy
Coursfy

Turn your expertise into a thriving online academy — courses, community, and payments in one place.

BROWSE

  • All academies
  • Free academies
  • By category
  • New this week

LEARNERS

  • Use an invite
  • Certificates
  • Help Center
  • FAQs

CREATORS

  • Request a call
  • White-label & SSO
  • Payouts & fees
  • Success stories

Company

  • About us
  • Blog
  • Careers
  • Contact Us

© 2026 Coursfy. All rights reserved.

Privacy PolicyTerms of ServiceCookie PolicyAccessibility Statement
Hi, I'm Lumina — your learning agent

Anything you need to know, as a course you can finish.

Tell Lumina what you want to learn. Get a course with clear lessons and interactive practice built around your goal.

Type what you want to learn, or pick one below

Explore exercises and features

There is more inside a Lumina lesson.

Explore features & exercises

Practise the skill

Solve problems, order steps, label diagrams and work through decisions.

See the idea move

Explore charts, timelines, code traces, graphs, simulations and decision scorecards.

Ask about the part you are learning

Keep the lesson beside your conversation. Highlight a passage for help, ask a follow-up and keep your notes in the same workspace.

Come back to what matters

Use flashcards and lesson notes to revisit ideas after your first pass.

How it works

Three steps, and the middle one isn't yours.

Pick a step to see the screen it produces. The answers you give below actually change the outline.

Step 1 of 2 · Shape the courseLive — try it
Your goal
Understand how neural networks actually learn, enough to read a paper
Where are you starting from?
This changes the first module, not the destination.
How much time per week?
Lumina sizes the lessons to fit — lessons land at 15–18 minutes
Inside a lesson

Not a wall of text with a quiz stapled on.

Every lesson has the same five parts. You can't scroll past the check — it has to be answered before the next lesson opens.

1
Reading, one idea at a time

Long enough to teach one idea properly, written for the level you said you were at.

2
One figure that earns its place

A diagram or chart only where the words are doing badly. No decorative illustration.

3
An exercise matched to the subject

Code that runs, a figure you label, a call you have to keep alive. Chosen by subject, not by what's cheap to grade.

4
A check you can't scroll past

One question that has to be answered before the next lesson unlocks. Wrong answers get told why.

5
The tutor, already in context

It has this lesson, your answers and your uploaded material open. No re-explaining yourself.

07
Why the gradient shrinks
Lesson 7 of 12 · 18 min
1 · Reading

Each layer's gradient is a product of the ones above it. Multiply enough numbers below one together and you approach zero — which is why a deep sigmoid network stops learning at the bottom long before it stops learning at the top.

2 · One figure
Gradient magnitude by layer, input on the left.
3 · Exercise
Spot the flawpicked for this subject
This training curve flattens at epoch 40. Two explanations are on the table — say which one the curve rules out.
4 · The check you can't skip
5 · Stuck on this? Ask me — I have the lesson and your material open.
EXPLORE LUMINA

Find a different way to practise.

From a quick recall question to a case that unfolds with your decisions. Explore the formats Lumina can bring into a lesson.

5 formats to explore

  • PRACTICE

    Explore code

    Read a program and reason about its output. Editing and execution are available when the course’s code sandbox is enabled.

  • PRACTICE

    Live working

    Build the answer line by line; each line is marked.

  • PRACTICE

    Order it

    Put the steps in the right order against a constraint.

  • PRACTICE

    Spot the flaw

    Find what is wrong in a passage and name the issue.

  • PRACTICE

    Work in the sheet

    Type formulas into cells; the formula is marked, not the number.

Your course uses formats that fit its subject and learning goals. Available activities and AI help can vary by course and plan.

Explore features & exercises
The tutor

It already knows which lesson you're on.

No re-explaining your context in a fresh chat. The tutor has the lesson, your answers, the exercise you just got wrong, and any material you uploaded — and it points back at the lesson rather than inventing a new one.

Answers point back at the lesson, and say which one.
It can see the exercise you just failed, and marks from there.
Part of Coursfy Pro, within a daily AI allowance that resets at midnight UTC.
LuminaLesson 7 openIncluded with Coursfy Pro
Why does the gradient shrink in the early layers?
Because each layer's gradient is a product of the ones above it. In the figure in this lesson, every sigmoid derivative is at most 0.25 — multiply four of those together and you're at 0.004 before the weights are even involved. The early layers aren't learning slowly because they're less important; they're learning slowly because the number reaching them is tiny.
From lesson 7, the figure you just read
Ask something else
Exam practice

Preparing for an exam?

Name the exam. Lumina maps its topics, gives you a 20-question diagnostic, shows your score topic by topic, builds a short course on your weakest topic and times practice exams.

Start exam practice

Practice questions written by Lumina, not real exam questions. Not affiliated with any exam organisation.

Material to build on
PDF
deep-learning-notes-ch1-4.pdf
62 pages · indexed, 4 chapters recognised
Ready
DOCX
backprop-worked-examples.docx
12 pages · indexed, 4 chapters recognised
Ready
XLSX
training-runs-learning-rates.xlsx
3 sheets · indexed
Ready
Add a PDF, spreadsheet or document
Your own material

Build the course on your syllabus, not the internet's.

Drop in the lecture PDF, the internal handbook or the spreadsheet. Lumina reads it once and builds the lessons around what's actually in it — and the course shows which file it was built from.

The course shows which file it was built from.
Uploaded once, reused across every course you build on it.
Your files stay yours — they don't train anything.
Why not just ask a chatbot

You can. You just won't finish.

A chat window is excellent at answering the question in front of it. It isn't a course, and after three weeks you have a scrollback instead of a skill.

A chat window
A Lumina course
StructureYou have to steer, every single sessionModules in an order, with a first and a last lesson
PracticeYou have to think of the exercises yourselfAn exercise style chosen for the subject, and graded
Being wrongTends to agree with youMarks per line and won't unlock the next lesson
ProgressNothing is trackedLessons done, streak, and where you stopped
Your materialPasted in again each sessionRead once, used in every lesson
After three weeksA long scrollbackA finished course, flashcards and a certificate
At the end

A certificate that says what you actually did.

Not a completion badge for scrolling to the bottom. It records the modules you finished, the exercises you passed and the date — with a verification link anyone can check.

Lists the modules completed and exercises passed, not just a date.
A verification link that works without an account.
Free courses earn one too — the certificate isn't the paywall.
See how verification works
Coursfy
Sample
Built with Lumina
Certificate of completion
How neural networks actually learn
Awarded to the learner who finished it, on the day they finished.
12 / 12
Lessons
94%
Checks passed
3h 20m
Time on task
Every certificate carries a code. Verifiable without an account.
Who uses it

Same engine, four different reasons.

Learners
Learn the thing no one has made a course about

Your goal is too specific for a catalogue — a niche exam, an internal tool, one paper you need to understand. Say it and start.

Build a course
Teams & L&D
Onboarding built on your own handbook

Upload the internal material, generate role-specific courses, and hand them out with the invite.

Ask for a quote
Creators
Draft in minutes, then make it yours

Lumina writes the first pass of an academy course. You rewrite the parts only you could teach and keep the structure.

For creators
Schools
One goal, a class-sized course

Set the goal, build the course in your academy, invite the class and follow each learner's progress. Learners never see a card.

Talk to us
Plans

Start free, upgrade when you want more.

Pro adds a monthly credit allowance and any course size.

Free
€0to start

Enough to see whether a Lumina course actually works for you.

Take a course from start to finish, certificate included
No card required
Start learning
Most learners
Coursfy Pro

Pro pricing isn't available yet. Check back soon.

Everything in Free
150 credits a month — a module costs from 1 credit
Longer generated courses, with more modules
The AI tutor inside every lesson
Your PDF, Excel and Word files as source material
See plans
FAQ

Before you build one

It's judged on whether you can do the thing at the end. Every lesson ends in a check you can't skip. Where the subject can be practised, an exercise matched to it — code that runs, a figure you label, a call you have to keep alive. If a module misses, regenerate just that module.

The thing you've been meaning to learn is one sentence away.

No card to start. If the outline isn't right, regenerate a module or edit it yourself — it's your course either way.

Build my first courseSee my courses
Hi, I'm Lumina — your learning agent

Anything you need to know, as a course you can finish.

Tell Lumina what you want to learn. Get a course with clear lessons and interactive practice built around your goal.

Type what you want to learn, or pick one below

Explore exercises and features

There is more inside a Lumina lesson.

Explore features & exercises

Practise the skill

Solve problems, order steps, label diagrams and work through decisions.

See the idea move

Explore charts, timelines, code traces, graphs, simulations and decision scorecards.

Ask about the part you are learning

Keep the lesson beside your conversation. Highlight a passage for help, ask a follow-up and keep your notes in the same workspace.

Come back to what matters

Use flashcards and lesson notes to revisit ideas after your first pass.

How it works

Three steps, and the middle one isn't yours.

Pick a step to see the screen it produces. The answers you give below actually change the outline.

Step 1 of 2 · Shape the courseLive — try it
Your goal
Understand how neural networks actually learn, enough to read a paper
Where are you starting from?
This changes the first module, not the destination.
How much time per week?
Lumina sizes the lessons to fit — lessons land at 15–18 minutes
Inside a lesson

Not a wall of text with a quiz stapled on.

Every lesson has the same five parts. You can't scroll past the check — it has to be answered before the next lesson opens.

1
Reading, one idea at a time

Long enough to teach one idea properly, written for the level you said you were at.

2
One figure that earns its place

A diagram or chart only where the words are doing badly. No decorative illustration.

3
An exercise matched to the subject

Code that runs, a figure you label, a call you have to keep alive. Chosen by subject, not by what's cheap to grade.

4
A check you can't scroll past

One question that has to be answered before the next lesson unlocks. Wrong answers get told why.

5
The tutor, already in context

It has this lesson, your answers and your uploaded material open. No re-explaining yourself.

07
Why the gradient shrinks
Lesson 7 of 12 · 18 min
1 · Reading

Each layer's gradient is a product of the ones above it. Multiply enough numbers below one together and you approach zero — which is why a deep sigmoid network stops learning at the bottom long before it stops learning at the top.

2 · One figure
Gradient magnitude by layer, input on the left.
3 · Exercise
Spot the flawpicked for this subject
This training curve flattens at epoch 40. Two explanations are on the table — say which one the curve rules out.
4 · The check you can't skip
5 · Stuck on this? Ask me — I have the lesson and your material open.
EXPLORE LUMINA

Find a different way to practise.

From a quick recall question to a case that unfolds with your decisions. Explore the formats Lumina can bring into a lesson.

5 formats to explore

  • PRACTICE

    Explore code

    Read a program and reason about its output. Editing and execution are available when the course’s code sandbox is enabled.

  • PRACTICE

    Live working

    Build the answer line by line; each line is marked.

  • PRACTICE

    Order it

    Put the steps in the right order against a constraint.

  • PRACTICE

    Spot the flaw

    Find what is wrong in a passage and name the issue.

  • PRACTICE

    Work in the sheet

    Type formulas into cells; the formula is marked, not the number.

Your course uses formats that fit its subject and learning goals. Available activities and AI help can vary by course and plan.

Explore features & exercises
The tutor

It already knows which lesson you're on.

No re-explaining your context in a fresh chat. The tutor has the lesson, your answers, the exercise you just got wrong, and any material you uploaded — and it points back at the lesson rather than inventing a new one.

Answers point back at the lesson, and say which one.
It can see the exercise you just failed, and marks from there.
Part of Coursfy Pro, within a daily AI allowance that resets at midnight UTC.
LuminaLesson 7 openIncluded with Coursfy Pro
Why does the gradient shrink in the early layers?
Because each layer's gradient is a product of the ones above it. In the figure in this lesson, every sigmoid derivative is at most 0.25 — multiply four of those together and you're at 0.004 before the weights are even involved. The early layers aren't learning slowly because they're less important; they're learning slowly because the number reaching them is tiny.
From lesson 7, the figure you just read
Ask something else
Exam practice

Preparing for an exam?

Name the exam. Lumina maps its topics, gives you a 20-question diagnostic, shows your score topic by topic, builds a short course on your weakest topic and times practice exams.

Start exam practice

Practice questions written by Lumina, not real exam questions. Not affiliated with any exam organisation.

Material to build on
PDF
deep-learning-notes-ch1-4.pdf
62 pages · indexed, 4 chapters recognised
Ready
DOCX
backprop-worked-examples.docx
12 pages · indexed, 4 chapters recognised
Ready
XLSX
training-runs-learning-rates.xlsx
3 sheets · indexed
Ready
Add a PDF, spreadsheet or document
Your own material

Build the course on your syllabus, not the internet's.

Drop in the lecture PDF, the internal handbook or the spreadsheet. Lumina reads it once and builds the lessons around what's actually in it — and the course shows which file it was built from.

The course shows which file it was built from.
Uploaded once, reused across every course you build on it.
Your files stay yours — they don't train anything.
Why not just ask a chatbot

You can. You just won't finish.

A chat window is excellent at answering the question in front of it. It isn't a course, and after three weeks you have a scrollback instead of a skill.

A chat window
A Lumina course
StructureYou have to steer, every single sessionModules in an order, with a first and a last lesson
PracticeYou have to think of the exercises yourselfAn exercise style chosen for the subject, and graded
Being wrongTends to agree with youMarks per line and won't unlock the next lesson
ProgressNothing is trackedLessons done, streak, and where you stopped
Your materialPasted in again each sessionRead once, used in every lesson
After three weeksA long scrollbackA finished course, flashcards and a certificate
At the end

A certificate that says what you actually did.

Not a completion badge for scrolling to the bottom. It records the modules you finished, the exercises you passed and the date — with a verification link anyone can check.

Lists the modules completed and exercises passed, not just a date.
A verification link that works without an account.
Free courses earn one too — the certificate isn't the paywall.
See how verification works
Coursfy
Sample
Built with Lumina
Certificate of completion
How neural networks actually learn
Awarded to the learner who finished it, on the day they finished.
12 / 12
Lessons
94%
Checks passed
3h 20m
Time on task
Every certificate carries a code. Verifiable without an account.
Who uses it

Same engine, four different reasons.

Learners
Learn the thing no one has made a course about

Your goal is too specific for a catalogue — a niche exam, an internal tool, one paper you need to understand. Say it and start.

Build a course
Teams & L&D
Onboarding built on your own handbook

Upload the internal material, generate role-specific courses, and hand them out with the invite.

Ask for a quote
Creators
Draft in minutes, then make it yours

Lumina writes the first pass of an academy course. You rewrite the parts only you could teach and keep the structure.

For creators
Schools
One goal, a class-sized course

Set the goal, build the course in your academy, invite the class and follow each learner's progress. Learners never see a card.

Talk to us
Plans

Start free, upgrade when you want more.

Pro adds a monthly credit allowance and any course size.

Free
€0to start

Enough to see whether a Lumina course actually works for you.

Take a course from start to finish, certificate included
No card required
Start learning
Most learners
Coursfy Pro

Pro pricing isn't available yet. Check back soon.

Everything in Free
150 credits a month — a module costs from 1 credit
Longer generated courses, with more modules
The AI tutor inside every lesson
Your PDF, Excel and Word files as source material
See plans
FAQ

Before you build one

It's judged on whether you can do the thing at the end. Every lesson ends in a check you can't skip. Where the subject can be practised, an exercise matched to it — code that runs, a figure you label, a call you have to keep alive. If a module misses, regenerate just that module.

The thing you've been meaning to learn is one sentence away.

No card to start. If the outline isn't right, regenerate a module or edit it yourself — it's your course either way.

Build my first courseSee my courses