Learning LangChain

说明与用途

作者展示了如何将手工构建的问答管道迁移到 LangChain，利用其组件和管道语法实现更简洁的构建、可视化和性能提升。

问题

I had built a custom question‑answering app (Synapse) by hand and wanted to know what LangChain adds beyond what I already did.

结果

LangChain provided a standard shape for each pipeline step and a pipe syntax to compose them, reducing boilerplate and enabling features like graph visualization, parallel execution, and batching.

工作流

• Load PDFs and Word docs with LangChain loaders
• Split documents by headings and size (≈500 chars)
• Embed chunks with OpenAI embeddings (1536‑dim vectors)
• Store vectors in Chroma on disk
• Retrieve relevant chunks via similarity or MMR
• Prompt LLM with context and question using ChatPromptTemplate
• Parse LLM output with StrOutputParser
• Invoke chain with a question

证据

I built Synapse, a small app that answers questions using my own notes. I did every part of it by hand: load the files, turn them into vectors, find the closest ones to the question, and hand those to an LLM.

来源

LangChain gives each of those its own class. Nothing magic, it's the same thing the raw API wants.

来源

The pipe comes in one chain: chain = ({"context": retriever, "question": RunnablePassthrough()}) | prompt_template | chat | StrOutputParser()

来源