Example code for building applications with LangChain.js, with an emphasis on more applied and end-to-end examples than contained in the main documentation.
These cookbooks are in Jupyter notebook form and use the Deno runtime and the experimental Deno Jupyter Kernel (requires >= Deno v1.37).
Full installation instructions are available here: https://docs.deno.com/runtime/manual/tools/jupyter
Note that you will also need to install the Python jupyter package, and that the syntax for imports and environment variables are slightly different from Node and the web. In particular, we use Deno.env.get() to retrieve environment variables, and e.g. import { PromptTemplate } from "https://esm.sh/langchain/prompts"; to import from a URL to match Deno conventions.
| Notebook | Description |
|---|---|
| rewrite.ipynb | Handle real-world questions that contain extraneous, distracting information in your RAG chains by first rewriting them before performing retrieval. |
| rag_fusion.ipynb | Turn user queries into more search friendly queries, then query a vector store and use reciprocal rank fusion to rank the results. |
| basic_critique_revise.ipynb | Basic example of correcting an LLM's output using a pattern called critique-revise, where we highlight what part of the output is wrong and re-query the LLM for a correction. |
| step_back.ipynb | Example of a step back prompting technique, where we ask the LLM to take a step back and rephrase the original query for a more search friendly question. |