diff --git a/README.md b/README.md
index d6f03c9..00b8bfd 100644
--- a/README.md
+++ b/README.md
@@ -44,11 +44,21 @@ pip install -e .
 
 Or you can just copy `src/slope` to your projects.
 
-# Docs
+# Usage
 
-Docs are in [radenmuaz.github.io/slope-ad](https://radenmuaz.github.io/slope-ad)
+## Tutorials
 
-For a quick overview on how to train and backprop something, go to the [Quickstart Tutorial](./docs/tutorials/quickstart.md)
+[Quickstart Tutorial](./docs/tutorials/quickstart.md): How Tensors work, how to write and jit compile functions and train something
+
+[NN Training](./docs/tutorials/quickstart.md): Train a MLP on MNIST and ResNet on CIFAR-10
+
+[Internals Walkthrough](./docs/tutorials/internals_walkthrough.md): Understand the core of SlopeAD (hint: like JAX)
+
+[Extending SlopeAD](./docs/tutorials/internals_walkthrough.md): Add new backend, operators, procedures. Modify the core functions.
+
+## Docs and API reference
+
+Docs are available online at [radenmuaz.github.io/slope-ad](https://radenmuaz.github.io/slope-ad)
 
 # Features
 
@@ -76,11 +86,11 @@ For a quick overview on how to train and backprop something, go to the [Quicksta
         - Init: `full arange random_normal random_uniform`
         - GeneralReduce: `matmul conv gather_nd scatter_nd`
     - Composite operators system with "procedures" [slope/procedures.py](./src/slope/procedures.py)
-        - Procedures are functions containing calls to operators, exposed with `Tensor.procedure_name(*args)` syntax.
-        - Useful for definitions like:
+        - For defining Tensor functions composed with core operators, e.g.
           - `x.cos()`, where `def cos(x): return (math.pi/2 - x).sin()`
           - `x.conv_transpose(w)`: where `def conv_transpose(x, w): ...` is a very long function.
-    - An operator can be directly implemented as code translation to backend, or _fallback_ to a procedure, e.g. there is `conv` procedure in case the backend has no implementation for it.
+        - Procedures are functions containing calls to operators, exposed with `Tensor.procedure_name(*args)` syntax.
+        
 
 6. Extensible
     - Add new backend by defining implementation translations [slope/backends](./src/slope/backends)
diff --git a/docs/index.md b/docs/index.md
index 05be8eb..0da06ab 100644
--- a/docs/index.md
+++ b/docs/index.md
@@ -2,7 +2,7 @@
 
 Slope is a small automatic differentation (AD) engine, focused on machine learning (ML)
 
-This project is designed to be a hackable, educational AD engine focused on ML, yet able to do things end-to-end training to deployment, instead of just some simple toy examples.
+This project is designed to be a **hackable**, **educational** AD engine focused on ML, yet able to do things **end-to-end from training to deployment**, instead of just some simple toy examples.
 
 Tensor semantics are similar to Pytorch, functional API is similar to [JAX](https://github.com/google/jax), tensor operators code is heavily derived from [tinygrad](https://tinygrad.org/).
 
@@ -25,6 +25,26 @@ shape=(3,), dtype=float32, device='cpu:0'>
 ```
 
 
+# Features
+
+1. Functional API for forward-mode, reverse-mode, and higher-order AD, like in JAX:
+    - `grad vjp jvp jit vmap`
+    - `register_node tree_flatten tree_unflatten`
+
+3. Just-in-time compilation, code is compiled to one of these backends: [ONNX Runtime](https://onnxruntime.ai/), [OpenXLA IREE](https://iree.dev/) and NumPy.
+
+4. Training and inference, like [MLP on MNIST](examples/nn/mnist_mlp.py), [ResNet on CIFAR-10](examples/nn/cifar_resnet.py) and [export jitted function](examples/simple/export.py).
+
+5. Small (?), less than 3000 lines of core code [slope/core.py](./src/slope/core.py), after `black src --line-length 140`
+
+6. Operators and procedures system
+    - 33 core operators defined in [slope/operators.py](./src/slope/operators.py)
+    - `exp log sin sqrt invert cast stop_gradient add mul sub div pow equal less greater maximum sum max reshape expand permute slice pad flip cat full arange random_normal random_uniform matmul conv gather_nd scatter_nd`
+    - Composite operators system with "procedures" [slope/procedures.py](./src/slope/procedures.py), for defining functions like `cos`, `conv` and `avgpool2d` as functions calling core operators.
+
+7. Extensible, by writing new backend by defining implementation translations [slope/backends](./src/slope/backends), and adding more modules using [slope/nn.py](./src/slope/nn.py)
+
+
 # Install
 
 ## Stable release
@@ -43,45 +63,23 @@ pip install -e .
 
 Or you can just copy `src/slope` to your projects.
 
-# Docs
+# Usage
 
-Head over to [quickstart](./tutorials/quickstart.md)
+## Tutorials
 
-# Features
+[Quickstart Tutorial](./docs/tutorials/quickstart.md): How Tensors work, how to write and jit compile functions and train something
 
-1. Forward-mode, reverse-mode, and higher-order AD.
+[NN Training](./docs/tutorials/quickstart.md): Train a MLP on MNIST and ResNet on CIFAR-10
 
-2. Just-in-time compilation, with interchangeable backends, supporting CPU, CUDA and Metal:
-    - [IREE](https://iree.dev/) (StableHLO MLIR)
-    - [ONNX Runtime](https://onnxruntime.ai/) (ONNX)
-    - NumPy (CPU-only)
+[Internals Walkthrough](./docs/tutorials/internals_walkthrough.md): Understand the core of SlopeAD (hint: like JAX)
 
-3. Training and inference, examples:
-    - [MLP on MNIST](examples/nn/mnist_mlp.py)
-    - [ResNet on CIFAR-10](examples/nn/cifar_resnet.py)
-    - [Export jitted function](examples/simple/export.py)
+[Extending SlopeAD](./docs/tutorials/internals_walkthrough.md): Add new backend, operators, procedures. Modify the core functions.
+
+## Docs and API reference
+
+Docs are available online at [radenmuaz.github.io/slope-ad](https://radenmuaz.github.io/slope-ad)
 
-4. Small (?)
-    - <3000 lines of core code [slope/core.py](./src/slope/core.py), after run with `black src --line-length 140`
 
-5. Operators and procedures system
-    - 33 core operators defined in [slope/operators.py](./src/slope/operators.py)
-        - Unary: `exp log sin sqrt invert cast stop_gradient`
-        - Binary: `add mul sub div pow equal less greater maximum`
-        - Reduce: `sum max`
-        - Shape: `reshape expand permute slice pad flip cat`
-        - Init: `full arange random_normal random_uniform`
-        - GeneralReduce: `matmul conv gather_nd scatter_nd`
-    - Composite operators system with "procedures" [slope/procedures.py](./src/slope/procedures.py)
-        - Procedures are functions containing calls to operators, exposed with `Tensor.procedure_name(*args)` syntax.
-        - Useful for definitions like:
-          - `x.cos()`, where `def cos(x): return (math.pi/2 - x).sin()`
-          - `x.conv_transpose(w)`: where `def conv_transpose(x, w): ...` is a very long function.
-    - An operator can be directly implemented as code translation to backend, or _fallback_ to a procedure, e.g. there is `conv` procedure in case the backend has no implementation for it.
-
-6. Extensible
-    - Add new backend by defining implementation translations [slope/backends](./src/slope/backends)
-    - NN module [slope/nn.py](./src/slope/nn.py)
 
 
 
diff --git a/docs/tutorials/ad_vmap_tree.md b/docs/tutorials/ad_vmap_tree.md
deleted file mode 100644
index e69de29..0000000