01
Step 1
Understand
Reads your topic or your own notes — PDF, slides, Word.
A personalised AI tutor for learning AI
Everyone's learning AI. Nobody learns it the same way.
Shikshak builds a learning path around you — from neural-net basics to transformers — teaches each concept on video, checks you understood, and adjusts the difficulty, explanation and practice the moment you're not.
- English & हिन्दी
- A learning path that adapts to you
- A real mentor when it matters
↓ Live lesson
Live lesson · OverfittingTeaching
A model learns patterns from examples…
Watch it pause
The problem
Thousands of AI courses. One pace for everyone. Nobody sees where you got lost.
AI is full of stacked ideas — gradients, loss, attention. Miss one and everything after it blurs. Videos don't notice. Chatbots wait to be asked. Learners who are lost don't know what to ask.
…
0
students per teacher in many Indian classrooms
0%
of learners hesitate to ask a question in class
0
seconds — how long Shikshak waits before checking in
Illustrative figures for the story; Shikshak's numbers are measured per learner.
How it teaches
A lesson that watches you learn.
Eight steps a great teacher does without thinking — now done for every learner, every time.
01
Step 1
Understand
Reads your topic or your own notes — PDF, slides, Word.
02
Step 2
Personalise
Builds a learning path for your level, goal and pace — and re-orders it as you progress.
03
Step 3
Explain
Writes a script grounded in your material, with real examples.
04
Step 4
Show
Turns it into a video: a teacher, a voice, a board.
05
Step 5
Pause
Stops mid-video and asks about what was just said.
06
Step 6
Check
Grades your own words — not just multiple choice.
07
Step 7
Adapt
Adjusts difficulty, re-explains differently, or adds targeted practice.
08
Step 8
Hand over
Still stuck? A real mentor is told, with your answers.
Quick check — just like a lesson
So what does Shikshak do when you get an answer wrong?
When you're stuck
It never gives up on you. It changes how it teaches.
Wrong once
Re-explains with a new example.
Wrong again
Rebuilds that part from scratch — simpler words, new picture.
Still stuck
Pauses the lesson and tells your mentor, with your answers.
To: Mentor
Asha is stuck on gradient descent — 4 tries. Her answers are attached.
— A real person
Your notes, not the internet's
Upload your chapter. It teaches from that.
Every explanation points back to the page and section it came from. If your notes don't cover something, it says so — it never pretends.
9.4 Gradient Descent
Gradient descent moves each weight a small step against the gradient… the size of that step is the learning rate.
From your notes · page 4 · 2.3 Gradient Descent
Your language
Learn in the language you think in.
Switch between English and हिन्दी anytime — the whole classroom changes with you.
Same lesson, two languages
English
- 1A model learns patterns from examples…
- 2…but it can learn the noise too.
- 3We call this overfitting.
हिन्दी
- १मॉडल उदाहरणों से पैटर्न सीखता है…
- २…पर वह शोर भी सीख सकता है।
- ३इसे ओवरफ़िटिंग कहते हैं।
Built to be trusted
2–3
checkpoints in every video
0
re-explanations before a mentor steps in
0%
sources cited on every grounded lesson
0
languages, switchable mid-lesson
Under the hood
Not a chatbot. A teaching system.
Retrieval over your own document
Hybrid search with page-level citations; weak matches are refused, not guessed.
A checkpoint engine
Questions placed inside the video, about only what was said so far.
An adaptation ladder
Re-explain → rebuild → hand to a human. Rules, not vibes.
Video generation
Script, voice, lip-synced teacher and a board — rendered per concept.
Resilient by design
Reconnects mid-lesson, never re-generates what you already watched.
Honest scoring
Grades from your answers, not from what a model says about them.
Gemini · FastAPI · hybrid RAG · Next.js