CS-405 · Weeks
1 · Linear Models
2 · Loss Functions
3 · Optimisation
7 · Backpropagation
8 · CNN Architectures
9 · Regularisation
Up next
6 of 12 weeks complete
CS-405 · Machine Learning
Lecture 7 · 58:12 · transcribed
…the gradient flows backwards, so each weight learns its share of the error…
Assignment 2
Backprop by hand · due Fri
34 / 60 submitted
Quiz 4 · auto-graded
Course tutor
groundedWhy does the loss plateau after epoch 40?
Your learning rate is likely too high near the minimum — Lecture 7 covers this at 12:40.
Can you solve Q3 of Assignment 2 for me?
graded work
I can't solve it — but the chain-rule walkthrough at 31:05 covers the method.
AI teaches alongside your faculty. 24/7.
Auto-transcription, AI study assistants, smart grading and plagiarism detection — intelligence in every learning moment.
Every course gets a tutor grounded in your own materials — it cites the lecture, and it won't solve graded work.
Answer · grounded in your materials
Backpropagation pushes the loss gradient backwards through each layer, so every weight learns its share of the error…
Solve Q3 of Assignment 2 for me
cites the lecture · never does the assignment
Submissions grade themselves, MCQs and code auto-graded, essays scored against your rubric for a human to confirm.
quiz4.responses
lab3.zip
essay.pdf
Grading · CS-405 · week 6
AI gradedQuiz 4 · MCQ
10 / 10 · auto-graded
Lab 3 · code
14 / 15 tests passed
Essay · ethics of AI
rubric suggests 8 / 10
Lectures transcribe themselves, chapter markers, timestamps and searchable text — fifteen minutes per hour of video.
lecture-07.mp4
58:12 · uploaded 09:14
1 hour of video
Transcript · Lecture 7
ready in 14 min · searchable
15 minutes per hour of video
Every submission checked against public sources and the class itself — semantic matches highlighted with their sources.
Similarity report
semantic match · not just strings
¶ 4 · 82% match · nature.com/articles/s41586
¶ 7 · 64% match · class submission · 2201489
Course setup in zero clicks. Literally.
Courses auto-created from ERP subjects — students enrolled, faculty assigned. Your teachers just add content.
Build the curriculum in blocks, reorder lessons, set prerequisites, unlock dates and completion rules — drag, drop, done.
Week 1 · Foundations
complete · 240 / 240
Week 2 · Regression
open · 182 watching
Week 3 · Optimisation
unlocks Oct 14 · quiz to complete
dragging
Week 4 · Neural Nets
locked · prereq Week 3
prerequisites · unlock dates · completion rules
Every format, one library, videos, PDFs, notebooks and SCORM — AI processes each one for how students actually learn.
lecture-08.mp4
unit-2-notes.pdf
lab-gradient.ipynb
safety-module.zip
Course library · CS-405
Processedlecture-08.mp4
transcribed · chaptered
unit-2-notes.pdf
indexed · searchable
lab-gradient.ipynb
executable · graded cells
safety-module.zip
SCORM · unpacked
Atlas creates the subject, and Learn creates the course — students enrolled, faculty assigned, gradebook linked. Nobody clicks anything.
Atlas ERP · subject
CS-405 · Machine Learning
Sem 5 · created 09:00
created in the ERP
Pegasus Learn · course
zero clicksCS-405 · Machine Learning
240 students enrolled
Dr. Mehra assigned
Gradebook linked
Module shell created
ready 09:00:04 · four seconds after the ERP
Grades flow everywhere. Insight flows back.
Continuous assessment flows into the ERP gradebook in real time — and the analytics catch struggling students weeks early.
One unified grade book, continuous assessment flows to the ERP, exam results flow back — both ways, in real time.
Learn · continuous
Atlas ERP · gradebook
CA marks →
← exam results
one unified grade book · real time
See who's falling behind before it's too late — watch time, completion and engagement, flagged weeks before the exam.
Dev Patel
38% engaged
watch time ↓ · 3 missed submissions
flagged before the midterm, not after
Engagement · CS-405
watch time · completion · submissions
Quizzes write themselves, generated from your lecture transcripts across Bloom's levels — faculty just review and publish.
the source material
Quiz 5 · draft
generated from Lecture 7 · awaiting approval
Define the learning rate.
RecallWhy does momentum dampen oscillation?
ApplyCritique early stopping vs. L2 for this loss curve.
AnalyseBloom's spread · publish when ready
Runs on your GPUs, or partner H100 clusters on demand — the platform routes every AI workload to the right hardware.
On campus
Your GPU cluster
8× A100 · data stays home
Partner
H100 clusters
on demand · burst only
Tutor sessions
1,412 today
campus GPUTranscription queue
6 lectures
campus GPUQuiz generation
3 drafts
partner H100Semester model fine-tune
nightly
partner H100the platform routes every workload · you never think about it
AI
Quiz Generation
Questions from lecture transcripts at every difficulty level, using Bloom's taxonomy.
AI
Content Summarization
Chapter summaries, flashcards and study notes generated from uploaded materials.
AI
Subtitle Translation
Subtitles in Hindi, Tamil, Telugu and more — 5 minutes per hour of video per language.
AI
Adaptive Learning
Next-content recommendations from each student's completion pattern and performance.
Give every student a teacher that never sleeps.
Courses that build themselves from your ERP, AI that tutors from your own materials, and grades that flow everywhere — automatically.
