some docs refactoring

CHANGELOG.mdx

@@ -1827,10 +1827,10 @@ Upgrade dependent libraries with security vulnerabilities (Pillow, TensorFlow, u
 
 ### Bugfixes
 
-- [#8364](https://github.com/rasahq/rasa/issues/8364): Fixed a bug where [`ListSlot`](./concepts/domain.mdx#list-slot)s were filled with single items
+- [#8364](https://github.com/rasahq/rasa/issues/8364): Fixed a bug where [`ListSlot`](./building-classic-assistants/domain.mdx#list-slot)s were filled with single items
   in case only one matching entity was extracted for this slot.
 
-  Values applied to [`ListSlot`](./concepts/domain.mdx#list-slot)s will be converted to a `List`
+  Values applied to [`ListSlot`](./building-classic-assistants/domain.mdx#list-slot)s will be converted to a `List`
   in case they aren't one.
 
 - [#8581](https://github.com/rasahq/rasa/issues/8581): Fix bug with false rule conflicts
@@ -2547,7 +2547,7 @@ Upgrade dependent libraries with security vulnerabilities (Pillow, TensorFlow, u
   and continue finetuning the model on the complete dataset consisting of
   new training examples. To do so, use `rasa train --finetune`. For more
   detailed explanation of the command, check out the docs on [incremental
-  training](./concepts/command-line-interface.mdx#incremental-training).
+  training](./command-line-interface.mdx#incremental-training).
 
   Added a configuration parameter `additional_vocabulary_size` to
   [`CountVectorsFeaturizer`](./building-classic-assistants/components.mdx#countvectorsfeaturizer)
@@ -3267,7 +3267,7 @@ Upgrade dependent libraries with security vulnerabilities (Pillow, TensorFlow, u
 
 - [#4536](https://github.com/rasahq/rasa/issues/4536): Add support for Python 3.8.
 - [#5368](https://github.com/rasahq/rasa/issues/5368): Changed the project structure for Rasa projects initialized with the
-  [CLI](./concepts/command-line-interface.mdx) (using the `rasa init` command):
+  [CLI](./command-line-interface.mdx) (using the `rasa init` command):
   `actions.py` -> `actions/actions.py`. `actions` is now a Python package (it contains
   a file `actions/__init__.py`). In addition, the `__init__.py` at the
   root of the project has been removed.
@@ -3832,7 +3832,7 @@ Upgrade dependent libraries with security vulnerabilities (Pillow, TensorFlow, u
 
 - [#5521](https://github.com/rasahq/rasa/issues/5521): Responses used in ResponseSelector now support new lines with explicitly adding `\\n` between them.
 
-- [#5758](https://github.com/rasahq/rasa/issues/5758): Fixed a bug in [`rasa export`](./concepts/command-line-interface.mdx#rasa-export)) which caused Rasa Open Source to only migrate conversation events from the last [Session configuration](./concepts/domain.mdx#session-configuration).
+- [#5758](https://github.com/rasahq/rasa/issues/5758): Fixed a bug in [`rasa export`](./command-line-interface.mdx#rasa-export)) which caused Rasa Open Source to only migrate conversation events from the last [Session configuration](./concepts/domain.mdx#session-configuration).
 
 ## [1.10.1] - 2020-05-15
 
@@ -4253,7 +4253,7 @@ This can help to fix problems when using `rasa shell` with custom actions which
   Language models currently supported: BERT, OpenAIGPT, GPT-2, XLNet, DistilBert, RoBERTa.
 
 - [#5225](https://github.com/rasahq/rasa/issues/5225): Added a new CLI command `rasa export` to publish tracker events from a persistent
-  tracker store using an event broker. See [Export Conversations to an Event Broker](./concepts/command-line-interface.mdx#rasa-export), [Tracker Stores](./production/tracker-stores.mdx)
+  tracker store using an event broker. See [Export Conversations to an Event Broker](./command-line-interface.mdx#rasa-export), [Tracker Stores](./production/tracker-stores.mdx)
   and [Event Brokers](./production/event-brokers.mdx) for more details.
 
 - [#5230](https://github.com/rasahq/rasa/issues/5230): Refactor how GPU and CPU environments are configured for TensorFlow 2.0.
@@ -5067,9 +5067,7 @@ Regression: changes from `1.2.12` were missing from `1.4.0`, readded them
 - `FallbackPolicy` can now be configured to trigger when the difference between confidences of two predicted intents is too narrow
 
 - experimental training data importer which supports training with data of multiple
-  sub bots. Please see the
-  [docs](./concepts/training-data-importers.mdx) for more
-  information.
+  sub bots. Please see the docs for more information.
 
 - throw error during training when triggers are defined in the domain without
   `MappingPolicy` being present in the policy ensemble

docs/docs/action-server/validation-action.mdx

@@ -23,7 +23,7 @@ can be set or updated outside of a form context.
 :::note
 `ValidationAction` is intended for extracting slots **outside** the context of a form.
 It will ignore extraction and validation methods for any slots that have a form specified under the
-[slot mapping's `conditions`](../concepts/domain.mdx#mapping-conditions).
+[slot mapping's `conditions`](../building-classic-assistants/domain.mdx#mapping-conditions).
 It will not run these methods when the specified form is active nor when no form is active.
 Please extend the [`FormValidationAction`](./validation-action.mdx#formvalidationaction-class) class to apply
 custom slot mappings only within the context of a form.

docs/docs/building-classic-assistants/business-logic.mdx

@@ -35,7 +35,7 @@ The examples on this page come from the [formbot example bot](https://github.com
 ## Step-by-step Guide on Using Forms to Handle Business Logic
 
 [Forms](./forms.mdx) work by prompting the user for information until it has gathered all required information.
-The information is stored in [slots](../concepts/domain.mdx#slots). Once all the required slots are filled,
+The information is stored in [slots](./domain.mdx#slots). Once all the required slots are filled,
 the bot fulfills the user's original request.
 
 ### 1. Defining the form
@@ -65,7 +65,7 @@ forms:
         - outdoor_seating
 ```
 
-These slots need to be added to your domain's `slots` section, along with [slot mappings](../concepts/domain.mdx#slot-mappings) which define how the slots can be filled. For any slot filled `from_entity`, the entity also needs to be added
+These slots need to be added to your domain's `slots` section, along with [slot mappings](./domain.mdx#slot-mappings) which define how the slots can be filled. For any slot filled `from_entity`, the entity also needs to be added
 to the domain. Slots filled by forms should usually not influence
 the conversation, so set `influence_conversation` to `false:
 
@@ -122,7 +122,7 @@ filled based on the user's intent: If it is `affirm`, it'll be `true`, if it is
 However, the slot should only be set to `true` or `false` if the user was responding to the question, `Do you want to sit outside?`.
 To enforce this condition, the `conditions` for the `outdoor_seating` slot requires that `restaurant_form` is active and that the requested slot is `outdoor_seating`.
 If there were no conditions and the user had sent a message with the `affirm` or `deny` intent earlier in the conversation,
-the `outdoor_seating` slot would already be filled when the form was activated. Therefore the form would not prompt the user for their outdoor seating preference. See [mapping conditions](../concepts/domain.mdx#mapping-conditions) for more information.
+the `outdoor_seating` slot would already be filled when the form was activated. Therefore the form would not prompt the user for their outdoor seating preference. See [mapping conditions](./domain.mdx#mapping-conditions) for more information.
 
 #### Validating Slots
 
@@ -135,7 +135,7 @@ about validation actions.
 #### Requesting Slots
 
 To specify how the bot should ask for the required information,
-you define [`responses`](../concepts/domain.mdx#responses) called `utter_ask_{slotname}` in your domain:
+you define [`responses`](./domain.mdx#responses) called `utter_ask_{slotname}` in your domain:
 
 ```yaml-rasa title="domain.yml"
 responses:
@@ -149,7 +149,7 @@ responses:
 
 ### 2. Updating the configuration
 
-A form's [happy path](../concepts/glossary.mdx#happy--unhappy-paths) should be defined as a [rule](rules.mdx) which means you'll need to add the [RulePolicy](./policies.mdx#rule-policy)
+A form's [happy path](./glossary.mdx#happy--unhappy-paths) should be defined as a [rule](rules.mdx) which means you'll need to add the [RulePolicy](./policies.mdx#rule-policy)
 to your policies:
 
 ```yaml-rasa title="config.yml"

docs/docs/building-classic-assistants/chitchat-faqs.mdx

@@ -5,7 +5,7 @@ title: Chitchat and FAQs
 abstract: FAQ assistants are the simplest assistants to build and typically the first kind of assistant anyone builds. This page is a guide to the concepts and training data you need to handle non-contextual questions like FAQs and chitchat.
 ---
 
-[FAQs](../concepts/glossary.mdx#faqs) and [chitchat](../concepts/glossary.mdx#chitchat) are two cases where the conversational assistant responds with a fixed set of messages, and the assistant should always answer the same way,
+[FAQs](./glossary.mdx#faqs) and [chitchat](./glossary.mdx#chitchat) are two cases where the conversational assistant responds with a fixed set of messages, and the assistant should always answer the same way,
 no matter what has happened previously in the conversation.
 For example, in the following conversation, every question can be asked at any point in the conversation,
 with the answer being independent of anything the user has said previously.
@@ -82,7 +82,7 @@ pipeline:
 
 Consider an example where you have 20 different FAQs. Although each question is represented as an individual intent, all FAQ intents are handled the same way in the dialogue. For each FAQ intent, the assistant **retrieves** the proper response depending on which question has been asked.
 
-Instead of writing 20 rules, you can use a single action, e.g. `utter_faq` to handle all FAQs with a single rule by grouping them together under a single [retrieval intent](../concepts/glossary.mdx#retrieval-intent)
+Instead of writing 20 rules, you can use a single action, e.g. `utter_faq` to handle all FAQs with a single rule by grouping them together under a single [retrieval intent](./glossary.mdx#retrieval-intent)
 called e.g. `faq`.
 
 The single action uses the output of the
@@ -144,7 +144,7 @@ intents:
 
 Responses for the ResponseSelector follow the same naming convention as
 retrieval intents. Besides this, they can have all the characteristics of
-normal bot [response](../concepts/domain.mdx#responses). For the chitchat intents
+normal bot [response](./domain.mdx#responses). For the chitchat intents
 listed above, our responses could look like:
 
 ```yaml-rasa title="domain.yml"

docs/docs/building-classic-assistants/components.mdx

@@ -568,7 +568,7 @@ Note: The `feature-dimension` for sequence and sentence features does not have t
   **Configuring for incremental training**
 
   To ensure that `sparse_features` are of fixed size during
-  [incremental training](../concepts/command-line-interface.mdx#incremental-training), the
+  [incremental training](../command-line-interface.mdx#incremental-training), the
   component should be configured to account for additional patterns that may be
   added to the training data in future. To do so, configure the `number_additional_patterns`
   parameter while training the base model from scratch:
@@ -703,7 +703,7 @@ Note: The `feature-dimension` for sequence and sentence features does not have t
   **Configuring for incremental training**
 
   To ensure that `sparse_features` are of fixed size during
-  [incremental training](../concepts/command-line-interface.mdx#incremental-training), the
+  [incremental training](../command-line-interface.mdx#incremental-training), the
   component should be configured to account for additional vocabulary tokens
   that may be added as part of new training examples in the future.
   To do so, configure the `additional_vocabulary_size` parameter while training the base model from scratch:
@@ -1401,6 +1401,68 @@ See [starspace paper](https://arxiv.org/abs/1709.03856) for details.
 
 </details>
 
+### LLMIntentClassifier
+
+<RasaLabsLabel />
+
+<RasaLabsBanner version="3.7.0b1" />
+
+The LLM-based intent classifier is a new intent classifier that uses large
+language models (LLMs) to classify intents. The LLM-based intent classifier
+relies on a method called retrieval augmented generation (RAG), which combines
+the benefits of retrieval-based and generation-based approaches. 
+
+During training, the classifier
+
+1. embeds all intent examples and 
+2. stores their embeddings in a vector store.
+
+During prediction the classifier
+
+1. embeds the current message and
+2. uses the embedding to find similar intent examples in the vector store.
+3. The retrieved examples are ranked based on similarity to the current message and
+4. the most similar ones are included in an LLM prompt. The prompt guides the LLM to
+   predict the intent of the message.
+5. LLM predicts an intent label. 
+6. The generated label is mapped to an intent of the domain. The LLM can also
+   predict a label that is not part of the training data. In this case, the
+   intent from the domain with the most similar embedding is predicted.
+
+## Using the LLM-based Intent Classifier in Your Bot
+
+To use the LLM-based intent classifier in your bot, you need to add the
+`LLMIntentClassifier` to your NLU pipeline in the `config.yml` file.
+
+```yaml-rasa title="config.yml"
+pipeline:
+# - ...
+  - name: rasa_plus.ml.LLMIntentClassifier
+# - ...
+```
+
+## Customizing
+
+You can customize the LLM by modifying the following parameters in the
+`config.yml` file. **All of the parameters are optional.**
+See other LLM Components for how to customize the LLM parameters.
+
+### Fallback Intent
+
+The fallback intent is used when the LLM predicts an intent that wasn't part of
+the training data. You can set the fallback intent by adding the following
+parameter to the `config.yml` file.
+
+```yaml-rasa title="config.yml"
+pipeline:
+# - ...
+  - name: rasa_plus.ml.LLMIntentClassifier
+    fallback_intent: "out_of_scope"
+# - ...
+```
+
+
+
 ### FallbackClassifier
 
 - **Short**
@@ -2373,7 +2435,7 @@ Selectors predict a bot response from a set of candidate responses.
   It embeds user inputs and response labels into the same space and follows the exact same
   neural network architecture and optimization as the [DIETClassifier](./components.mdx#dietclassifier).
 
-  To use this component, your training data should contain [retrieval intents](../concepts/glossary.mdx#retrieval-intent). To define these,
+  To use this component, your training data should contain [retrieval intents](./glossary.mdx#retrieval-intent). To define these,
   checkout [documentation on NLU training examples](./training-data-format.mdx#training-examples) and
   [documentation on defining response utterances for retrieval intents](../concepts/responses.mdx#defining-responses).
 
@@ -2644,5 +2706,5 @@ pipeline:
 - name: "sentiment.SentimentAnalyzer"
 ```
 
-See the [guide on custom graph components](../concepts/custom-graph-components.mdx) for a complete guide on custom components.
+See the [guide on custom graph components](../concepts/components/custom-graph-components.mdx) for a complete guide on custom components.
 Also be sure to read the section on the [Component Lifecycle](./tuning-your-model.mdx#component-lifecycle).