AI Apps with ChatGPT and LangChain: The Introduction

Explore how large language models use transformer architectures and attention to understand context, tokenize text, create embeddings and context vectors, and perform autoregressive generation.

Examine how the GPT series advances via pre-training and fine-tuning, from GPT-1 to GPT-4, detailing parameters, memory, and compute, alongside related models like DALL-E and Whisper.

Explore how developers use ChatGPT via the web interface and the OpenAI API, compare free and paid models, and master prompt development, context management, and plugins for building AI apps.

Explore how the ChatGPT API solves text analysis tasks with long chain wrappers, token counting, and cost tracking, including summarization, sentiment analysis, translation, and rephrasing.

Explore OpenAI API model variants, focusing on GPT text models and chat completions, with snapshots for repeatable results, token-based pricing, and rate limits.

Set up your OpenAI account and API access to run course examples, note free trial rate limits, and securely store your secret key with a dot env file.

Learn to interact with the OpenAI API using the LangChain library, configure API keys via environment variables, and use chat and text models with messages, noting cost differences.

Explore how input context enables in-context learning and action planning, using open and closed book prompts, zero/one/few-shot patterns, and tool-driven agents to plan actions.

Explore how LangChain encapsulates rest APIs into Python and JavaScript classes to accelerate AI app development with prompt templates, memory tools, and data indexing.

Explore prompt engineering for developers, mastering the temperature parameter, and crafting clear, concise prompts with tone, instruction, and output format to control responses.

Explore prompt techniques to improve output quality in chat models, including self-confidence directives and chain-of-thought prompts that encourage longer reasoning and reduce hallucinations through practical examples.

Explore how prompt templates in the long chain prompts module standardize repeated prompts by defining fixed instructions and input variables, enable formatting outputs, and support sentiment analysis and translation.

Post-process ChatGPT responses using long-chain and structured output parsers to format, extract, and convert natural language into Python values or dictionaries, with formatting instructions in prompts.

Explore how to build a simple chain by combining a prompt and a large language model, then extend it with an output parser and optional cost tracking.

Demonstrate applying moderation techniques to both input and output in ai apps, using OpenAI moderation endpoints to classify content into hate speech, harassment, self-harm, and other categories.

Build an automated ai app that analyzes web shop reviews using prompts, prompt templates, long chain chains, and output parsing to translate, detect language, assess sentiment, and extract product details.

Explore the difference between chat and QA modes, highlighting how a stateless chat builds conversation by interleaving user questions and AI answers, tracking tokens, and using the full conversation.

Explore how chat conversation memory works using a conversation buffer memory and a conversation chain with OpenAI's chat GPT 3.5 Turbo, memory prompts, and history handling.

Explore building a working chat bot UI in a jupyter notebook by creating a conversation chain with gradio or panel, including chat box and clear history.

Explore how chat memory grows and how to manage it with buffer window memory, token-based memory, and summary memory. Learn practical examples of memory limits and keeping relevant history.

Explore how to summarize long PDF documents beyond token limits using a map-reduce summarization pipeline with PDF loaders and custom prompts.

Discover how the refine chain builds a final summary by iteratively refining an existing summary across pages, contrasting with the map-reduce chain for large PDFs.

Load web page content with a long chain using the web base loader, split into 3,000-character or 2,000-token chunks, then refine a summarization chain.

Summarize YouTube video transcripts with a YouTube loader and transcript API. Address transcription errors and generate concise results using refine and summary chains for AI apps with ChatGPT and LangChain.

Explore naive question answering on a pdf document by scanning every page for answers, noting the time and cost, then consider a faster map reduce chain that selects relevant fragments.

Learn how semantic similarity search uses text embeddings to pick the most relevant pages from long documents, store them in a vector database, and rank results by relevance scores.

Build a qa system on a pdf using text embeddings and semantic search with a vector db; retrieve two relevant pages and generate answers via a retrieval chain.

Build a persistent vector database from the pandas user guide using 645 chunks and 384-d embeddings with hugging face, then persist to disk for retrieval-based QA.

Create a GPT-3.5 Turbo powered pandas user guide chatbot using a persistent Chroma vector DB with embeddings, memory, and a conversational retrieval chain to fetch the most relevant documents.

Build a vector DB with text embeddings from code, then apply a conversational retrieval chain using ChatGPT to answer questions about code, demonstrated with TikTokin.

Explore the limitations of vector DB–driven retrieval QA for reasoning, illustrated by a cow fodder example, and compare retrieval gaps to knowledge representation with graph queries and ontologies.

Discover how agents built with LangChain and React reason, plan actions, and use tools to solve tasks, employing tool decorators, prompts, and an agent executor to produce final answers.

Explore the inner workings of the ReAct agent in the LangChain ecosystem, loading tools and monitoring calls with Lang Smith for end-to-end traceability.

Explore how OpenAI functions enable tool calls in GPT-3.5 Turbo and GPT-4, building an OpenAI functions agent with tools like get weekday and DuckDuckGo search.

Add chat memory to an OpenAI functions agent by integrating a conversation buffer memory and tool calls (get weekday, Wikipedia) to preserve chat history across questions.

Leverage a vector db to augment parametric memory with custom knowledge from a pdf research paper, using embeddings and a conversational retrieval agent.

Set up a Bing search API account and environment variables to enable Bing in a LangChain notebook, then use the Bing API wrapper with GPT-4 or GPT-3.5 turbo.

Develop an action planning workflow using simulated tools to analyze top billing customers, extract CRM requirements, and determine which product to focus on, prioritizing the planner for reliability and performance.

Explore ad hoc data analysis with a pandas DataFrame agent on a CSV file, using the Spaceship Titanic dataset. Learn to run Python code, handle missing values, and visualize results.

Explore building an SQL database agent with LangChain and SQLite Sakila, enabling ad hoc analysis through natural language queries and an SQL toolkit.

Explore building agents with multiple tools and control tool selection via descriptions and prompts, avoiding unnecessary API key setups, using Bing, DuckDuckGo, Wikipedia, and a calculator that evaluates Python expressions.