Retrieval Augmented Generation (RAG) with LangChain
| Course Details | ||||||||
|---|---|---|---|---|---|---|---|---|
|
||||||||
| Languages | ||||||||
|
||||||||
| Outcomes | ||||||||
|
||||||||
| Price & Terms | ||||||||
|
||||||||
| Technical | ||||||||
|
||||||||
| Specification | ||||||||
|
||||||||
Build retrieval augmented generation pipelines in LangChain, improve splitting and retrieval, evaluate them with Ragas, and try graph RAG in Neo4j.
The limited knowledge of a large language model restricts how well it serves a specific use case. Retrieval augmented generation closes that gap by pulling relevant passages from external data into the prompt. You load and split documents with LangChain, embed and store them, retrieve the most relevant chunks, and score the answers that come back. A graph-based approach then stores entities and their relationships in Neo4j as an alternative to vector search.
What You Will Achieve
- Build a LangChain retrieval chain that answers questions using your own PDF and HTML documents
- Choose a splitting method that respects code syntax, token limits or shifts in subject
- Score a RAG pipeline for faithfulness and context precision with Ragas
- Convert unstructured text into a Neo4j graph and answer questions with validated Cypher queries
Program Overview
Building RAG Applications with LangChain
Sets up a basic pipeline that loads documents, splits and embeds them, stores them in a vector store and retrieves them for a chat model.
- Document loaders for PDF and HTML files
- Recursive text splitting
- Embeddings and vector storage
- LCEL retrieval chain and prompt
Improving the RAG Architecture
Refines each stage of the pipeline with code-aware loading, finer splitting, sparse retrieval and evaluation metrics.
- Loading and splitting Python code files
- Splitting by tokens and semantically
- Sparse versus dense retrieval with BM25
- Ragas metrics and string evaluation
Introduction to Graph RAG
Replaces vector storage with a graph database built from graph documents, then tunes how the graph is queried inside a retrieval chain.
- Creating graph documents from text
- Storing and querying a graph database
- Building the Graph RAG chain
- Filtering, Cypher validation and few-shot prompts
Certificate Awarded (Paid)
Job Outlook
These retrieval skills apply in software engineering and AI application development, wherever a chat application built on a large language model has to answer from external documents, code files or graph data.
Good for: Developers who already build with LangChain and want their LLM applications to answer from external documents, code files and graph data.
Keep in mind: The retrieval lessons center on sparse and dense search with BM25; MRR and self-query retrieval get no lesson of their own.
Bottom line: A practical next step for LangChain retrieval work that adds evaluation with Ragas and a graph-based pipeline in Neo4j queried with Cypher.
Frequently Asked Questions
How does graph RAG differ from vector RAG?
Vector RAG retrieves the stored chunks most semantically similar to the user input. Graph RAG turns text into nodes and relationships and retrieves information by querying that graph, which captures how entities in the documents relate to each other.
Which kinds of files are loaded in the lessons?
The first chapter loads PDF and HTML files. A later chapter loads code files and splits Python files so that the chunks follow the code's syntax.
How is a RAG pipeline evaluated here?
The evaluation lessons use the Ragas framework to measure context precision and faithfulness, and add a string evaluation of the answers the pipeline returns.
What role does LangChain play in a RAG system?
LangChain is a framework for applications powered by large language models. It gives one consistent syntax for connecting the parts of a RAG system, such as the model, prompt templates and vector databases.
User Reviews
Write your review
You must be logged in to post a review.
Reviews can only be posted by registered learners. Log in or create a free account, then you will return here to post.
There are no reviews yet.