LangChain

Limitations of LLM

Outdate knowledge

Inability to take action ( not able to search , calculate , lookups )

Hallucination risks : bịa chuyện

Lack of context : not able to memory about the previous conversation

Stochastic parrot

Produce convincing language but lack of tru comprehension of meaning behind of words

Thiếu sự hiểu biết thật sự , chỉ bắt trước ngôn ngữ của con người bẵng xác xuất thống kê chứ không có ý thức lý luận hay is định

Why need to use langchain

Basic LLM API usage ( connec to external lib )

Facilitates advanced interaction like conversational context and persistent througy agent and memory

Key component of Langchain

Chain ( chaining )

agent ( make decision )

Tool ( integrate with external parties )

Memory ( keep conversation )

Pipeline of chain

Loading documents

Embedding retrieval

Query LLM

Parsing output

Writing memory

LLM and chat model - Langchain

Provide interface to connect and query language model like gpt , support asynchrony streaming , batch request

Document loaders : Langchain

Ingest data source into document with text and metadata

Document transformer : Langchain

Splitting , combining , filtering documents

Text embedding

Create vector representations of text for semantic search

Vector store Langchain

Store document , retrieved embedding documents

Tools : Langchain

Connect with 3 party

Agent : Langchain

Global driven action

Memory : Langchain

Persistence conversation

The main application of embedding data

• Semantic search

• Recommendation

• Filter