LangChain Application Development

Practical Guide to Building AI Apps with LangChain

AI Programming & Machine Learning 4.6(59)3h 33m18 lectures2 sections

What you'll learn

Understand the basics of the OpenAI API and its advanced use
Build LangChain models with OpenAI and Gemini
Design Prompt Templates
Use Few-Shot Learning
Structure outputs with Output Parsing
Build LangChain chains
Add memory to applications
Use Hugging Face with LangChain
Summarize YouTube videos with language models

About this course

This course teaches LangChain application development, meaning how to build AI applications based on large language models (LLMs). It starts with an introduction, then the basics of the OpenAI API and its advanced methods, followed by LangChain models using OpenAI and using Gemini. After that you learn Prompt Templates, Few-Shot Learning, Output Parsing, LangChain chains, and memory in LangChain across two parts. The course then moves to Hugging Face: an introduction to it, using it with LangChain, its models, and integrating it with LangChain. It also includes a project that summarizes YouTube videos in two parts, followed by the projects used in the course. It suits programmers and people interested in AI applications, and basic knowledge of the Python language is preferred. By the end you will be able to build a simple application that connects language models to tools, structure its outputs, add memory to it, and use Hugging Face models inside your own project.

Expected outcomes

Ability to build practical AI applications using LangChain.
Implement real projects based on LLMs from start to deployment.
Use HuggingFace with LangChain in real applications.
Analyze and organize outputs using Output Parsing techniques.

Learner reviews

These reviews are from learners who bought the content and completed at least 50% of it — there are no reviews from anyone who hasn't gone through the material.

4.6

★★★★★

59 reviews

5★

4★

3★

2★

1★

Course content

Developing LangChain Applications

17 lectures

Course IntroductionOutlines the LangChain course roadmap: understanding Large Language Models, the OpenAI API and request cost, then Prompt Engineering, Output Parsers, Chains and types of Memory, ending with using Hugging Face and integrating it with LangChain.2:59
Basics of OpenAI APICovers installing the OpenAI library, creating a client with an API key, sending a message with system and user roles, how token-based pricing works, and extracting the reply from the model's response.10:57
Advanced OpenAI APIWraps the OpenAI API call in a single function taking the user message, system message, model name, temperature and max tokens, then tests it with instructions that change the reply's dialect.12:18
LangChain Models Using OpenAIShows how to use OpenAI models through LangChain: installing the library, setting temperature, building a prompt template that rewrites a complaint email in a chosen style, reading token metadata, and streaming the reply word by word.12:19
LangChain Models Using GeminiExplains using Google's Gemini models inside LangChain: installing the langchain-google-genai library, getting an API key, rerunning the same email-style template, and comparing reply quality after switching to a stronger model.9:26
Prompt TemplatesExplains LangChain prompt templates: defining a template with input variables such as style and email text, then filling them with format to get the final message and sending it to the model.9:36
Learning by Examples (Few‑Shot Learning)Explains few-shot learning: giving the model examples, such as a country and its capital, so it learns the desired format, then building a FewShotPromptTemplate from the examples, a base template and a suffix to ask about a new country.14:26
Output ParsingExplains output parsing: defining a ResponseSchema to specify the fields to extract from a customer review, such as gift, delivery time and price, so the model returns them as JSON or a dictionary.16:00
LangChain ChainsExplains LangChain chains: a basic chain linking the model to a prompt to suggest a product name, then sequential chains that pass each step's output to the next, such as summarizing, with named outputs for each chain.14:27
Memory in LangChain (Part One)Explains why the model doesn't remember earlier messages, then uses ConversationChain with ConversationBufferMemory to keep the whole conversation, and ConversationBufferWindowMemory to remember only the last few messages.14:32
Memory in LangChain (Part Two)Explains two LangChain memory types: ConversationTokenBufferMemory, which keeps the conversation up to a set token limit, and ConversationSummaryMemory, which stores a summary that updates with each new message.7:47
Introduction to Hugging FaceIntroduces the open-source Hugging Face platform and its free Transformer-based models, and tries pipeline tasks in Google Colab such as text generation with GPT-2 and text classification for sentiment and grammatical correctness.16:00
Using Hugging Face with LangChainApplies sequence-to-sequence tasks with Hugging Face pipelines, such as translating text from Arabic into another language and generating a question from a given context, while tuning parameters like maximum output length.13:41
Hugging Face ModelsExplains loading the model and the tokenizer separately in Hugging Face instead of using a pipeline, converting text into tokens, feeding them to the model, and decoding the output back into readable text.15:15
YouTube Video Summarization (Part One)Builds the first part of a YouTube video summarizer: extracting the video ID from the link, fetching the transcript with the youtube-transcript-api library, and joining the transcript segments into a single text.20:16
YouTube Video Summarization (Part Two)Completes the YouTube summarizer by splitting the long transcript into fixed-size character chunks, summarizing each chunk with a summarization model, combining the summaries, and running the whole flow from link to summary.15:41
Integrating Hugging Face with LangChainExplains integrating Hugging Face models with LangChain: loading a lightweight Alibaba model, wrapping it as a text-generation pipeline with max tokens and temperature, then linking it to a prompt template in a chain to write a recipe.7:48