LangChain in Action: Develop LLM-Powered Applications

Create a virtual environment with Python -m venv in a folder, activate it across Windows, Linux, or Mac, and install IPython kernel package to run Jupyter notebooks in VS Code.

Rename the sample .env file to .env, copy your OpenAI API key from platform.openai.com, and load it with python-dotenv using load_dotenv and find_dotenv.

Explore the chat completions API and how Lang Chain simplifies building systems with a model and messages in system, user, and assistant roles, and how to parse the response.

Install LangChain, use OpenAI and chat OpenAI models, and run single or batch requests with a simple predict interface. Build prompts from system and human messages and inspect generations.

Explore how LangChain 1.0 introduces the content blocks property, a cross-provider standard for message content stored as a list of dictionaries, including text blocks.

Explore prompts and prompt templates in LangChain, using dynamic system messages with input and output languages. Learn few-shot and zero-shot prompting and the prompt template class to build LLM-powered applications.

Learn the basics of prompt engineering, covering zero-shot, few-shot, and chain-of-thought prompting, with examples and guidance on when to apply each for sentiment classification tasks.

Explore few-shot prompting with LangChain by building a template that classifies text sentiment and extracts a one-word subject, using zero-shot and example-driven learning.

Explore chain of thought prompting to encode reasoning in prompts, using chain-of-thought prompts and few short prompt template to classify statements as positive, neutral, or negative, and compare prompt strategies.

learn how to compose multiple prompts into a single pipeline prompt to reuse parts for complex prompts, including chain-of-thought prompts, using introduction, example, and execution templates and the format method.

Serialize prompt templates with LangChain, save them as YAML or JSON, and load prompts from disk using input variables. Learn how to combine LMS and prompts into chains.

Explore chains as a synergy of language models and prompts, and learn the legacy method alongside the Lang Chain Expression Language for advanced chain construction.

Explore the basics of LM chains and output parsers in LangChain, combining a model with a prompt to classify text into sentiment, subject, and price, outputting JSON.

Learn to build and use LangChain output parsers with response schemas, including sentiment, subject, and price, converting results to a clean json via a structured output parser.

Master advanced LLMChain workflows with multiple inputs by using a dictionary to pass input and language, enabling language-specific jokes like a parrot in German.

Explore how to chain multiple language model steps with sequential chains in LangChain, passing outputs between templates to produce review, comment, summary, and German translation.

Explore router chains that use an LM to route to positive, neutral, or negative templates, invoking a multi prompt chain with tailored responses and default handling.

Explore LangChain callbacks to hook into model interaction stages for logging, monitoring, and streaming data; use pre-built or custom handlers, and track token usage and costs with open callback.

Explore memory in LangChain using the conversation buffer memory to retain chat history. Wire memory into a conversation chain to maintain context and enable meaningful follow-ups.

Demonstrate how conversation summary memory compresses long chats into summaries to save tokens and costs, using a max token limit and the conversation summary buffer memory with a chat class.

Build a real chat bot using streamlit and integrate a prompt memory with an LM chain. Reference unused classes at the top as hints for implementation.

Build a Streamlit chatbot with LangChain using a prompt template fed by history and human input, plus a custom streamlit chat memory and OpenAI integration, and run with streamlit.

Learn to use the vanilla OpenAI package for function calling with function definitions, enabling chat completions to return JSON responses.

Explore function calling in LangChain using long chain, open functions, and various definitions—pedantic class, dictionary, or direct functions—through a pizza cost example.

Explore implementing a pizza database with LangChain function calls, including get pizza info and add pizza, then compare with OpenAI function calls and retrieval augmented generation short rack.

LangChain updates function calling to a user-friendly tool calling approach. Decorators or pydantic models bind multiple tools to an llm for weather and seating queries.

Explore retrieval augmented generation with LangChain in action, and master data pipelines from loading sources to embedding and vector-store retrieval for building RAG apps.

Load data from a source with a text loader, split content into chunks using a recursive character text splitter (100 size, 20 overlap), enabling retrieval augmented generation and metadata-driven searches.

Explore embeddings as vector representations of semantic meaning in a vector space, create embeddings with the embeddings class using the order 002 model, and compare vectors with cosine similarity.

Load vectors into the files vector store, serialize with pickle, and retrieve documents with a retriever; then build a prompt with context to query the LM.

Learn to use LangChain's new runnable retrieval qa chain, migrate from the deprecated interface, and implement a flow with OpenAI llm, a context prompt, and create staff documents chain.

Build a real-world rag service with fastapi, a vector store, and a retrieval qa chain using few-shot prompting to respond in pirate language via a post endpoint and swagger docs.

Explore LangChain's high level agent, an LLM-powered system that reasons, plans, and executes tasks using tool calling, reflects on results, and dynamically selects tools to complete goals.

Learn to build high level agents with create_agent by wiring a model to tools, enabling checkpoint memory, adding middleware, and extending with custom state for personalized responses.

Discover how the indexing API keeps your vector store in sync by loading and updating documents with a record manager that hashes content, metadata, and source ids.

Install Docker Desktop, then use the Docker CLI to launch a Postgres service with a docker compose file in VSCode, and run docker compose up --build to pull images.

Set up Postgres with a dockerized vector service, expose port 5433, and initialize a vector store with a record manager to index documents.

Load and index documents into a vector store using the indexing API, exploring cleanup modes none, incremental, and full, with source metadata and record management.

Explore decoupled document retrieval with vector stores in LangChain, using the as retriever method to fetch relevant documents with similarity scores, and build production-ready, scalable applications.