2023-10-17 Chatbot Architecture

 

Services diagrams

Each at the end will be a separate GIT project

configures
configures
uses
uses
uses
uses
uses
api_chat_core
add_chatbot(api_key: str) : chatbot_id
add_fact(ChatFact: JSON, chatbot_id:int, api_key: str) : fact_id
update_fact(ChatFact: JSON, chatbot_id:int, api_key: str)
remove_fact(fact_id: int, chatbot_id:int, api_key: str)
list_facts(chatbot_id:int, api_key: str) : List[ChatFact]
remove_chatbot(chatbot_id:int, api_key: str)
new_chat(chatbot_id:int, api_key: str) : chat_id
configure_chat(language:EnumLangCode, SettingsChat, chatbot_id:int, api_key: str)
prompt(ChatPrompt:JSON, chatbot_id:int, api_key: str) : ChatHistory
api_llm
prompt(LLMPRompt:JSON, api_key: str) : LLMPromptResponse
api_text_embeddings
encode(EmbRequest:JSON, api_key: str) : Embedding
similarity(a: Embedding, b: Embedding, api_key: str) : float
api_documents_parser
parse_pdf(FilePDF, api_key: str) : List[ChatFact]
parse_url(FilePDF, api_key: str) : List[ChatFact]
frontend_chat
/home
/register
/login
/configure_chatbot
/configure_knowledge
/configure_integrations
/display_conversation_flow_tree
/display_web_chatbot
/update_knowledge_sources
integrations_chat
configure_zendesk(..)
configure_whatsapp(..)
configure_facebook(..)
api_conversation_tree
add_tree(api_key: str) : tree_id
add_dialog(ChatHistory, api_key: str, tree_id: int)
list_trees(api_key: str)
remove_tree(api_key: str, tree_id: int)
get_conversation_trees(filter_criteria, api_key: str, tree_id:int) : ConversationTree

 

APIs should be based on:

  1. FastAPI

  2. SwaggerUI for docs and testing

  3. Model View Controller

    1. Model classes as Python Dataclasses with ORM (SQLAlchemy) and PostgreSQL DB, Example: ^ All Data structures must be data classes not dictionaries

    2. View, FastAPI controller classes Example:

    3. Controller, Logic separated by classes

 

API - Chat core

Main task is to control core chat functionality, semantic search, prompt engineering.

For THE first version Embeddings, Similarity measurement and LLMS can be included in Chat Core API, but later should be improved independently of this node.

Plausible data structure

1 to N
1 to N
1 to N
1 to N
reused for all chats
clients
client_id: PK
api_key: sha1
created: timestamp
modified: timestamp
chatbots
chatbot_id: PK
client_id: FK
is_deleted: bool
created: timestamp
modified: timestamp
chatbot_languages
chatbot_language_id: PK
chatbot_id: FK
lang_code: str
created: timestamp
modified: timestamp
chatbot_facts
chatbot_fact_id: PK
chatbot_id: FK
text_segment_original: text
lang_code_original: str
text_segment_en: text
embedding_en: vector[256]
reference_text: text,
reference_url: text
is_deleted: bool
created: timestamp
modified: timestamp
chatbot_tags_in_facts
chatbot_tags_in_facts_id: PK
chatbot_tag_id: FK
chatbot_fact_id: FK
created: timestamp
modified: timestamp
chatbot_tags
chatbot_tag_id: PK
tag_name_en: str
tag_embedding_en: vector[256]
created: timestamp
modified: timestamp