Building Smarter Real-World Generative AI Systems

Explore the capabilities of generative AI, including text generation, image creation, music composition, and video generation. Discover the four-step process—data collection, training, pattern learning, and generation—that enables authentic content creation.

Explore large language models, their transformer architecture and billions of parameters, trained on diverse data, fine-tuning for translation, summarization, and question-and-answer tasks, plus diverse model types and access options.

Discover how Lang Chain links language models to build scalable, production-ready applications, using Lang Graph, integrations, Lang Graph Cloud, and Lang Smith for debugging and monitoring.

Explore the long chain dataflow framework, transforming raw input into meaningful results by routing through prompt templates, large language models or chat models, and the parse module to output.

Install Python on Windows using Anaconda and register Anaconda as the default Python version (3.12), then use Anaconda Navigator to manage environments, install packages, and launch Jupyter Notebook or Spyder.

Create and activate an Anaconda virtual environment, then install LangChain and LangGraph to build real-world generative AI applications.

Set up an OpenAI account with Google, access the API dashboard, and generate a secret API key, copying it securely for future integration of OpenAI services.

Define tools as modular components that agents call to execute actions or retrieve information, extending language models to interact with external systems and APIs.

Explore the types of tools and how predefined tools handle api requests, querying databases, and data processing. Learn to create custom tools using decorators or subclassing base tool classes.

Explore built-in tools like google search, google finance, and yahoo finance, with pricing that varies by usage; visit the Lang Chain website for details on other tools and pricing.

Explore the components of a tool: its unique identifier, name, and description; define the input schema with JSON, the function and optional async version, and the return type.

An agent augments a language model like GPT with reasoning, decision making, and action. It analyzes context, selects actions, and executes tasks beyond text generation.

Explore the react agent, an advanced agent that integrates reasoning and acting in a continuous loop, adapting in real time as Lang chain shifts to Lang graph.

Explore react agent in detail, showing how reasoning and acting enable AI agents to think through problems with internal dialogue and step-by-step reasoning, executing actions like code and API calls.

Examine the disadvantages of long chain agents, including setup complexity with multiple tools and APIs. Note resource intensity, slower performance, higher costs, and limited flexibility in dynamic situations.

Explore Landgraf, a stateful, multi-actor library for building intelligent applications with large language models and agent workflows. Scale complex systems with agent and multi-agent workflows, developed by Lange Chain, Inc.

Explore LangGraph's cycles and branching for loops and conditionals, plus pause and resume execution, human in the loop line graph, streaming outputs, and api webhooks to external systems.

Explore a real-world vacation planning scenario, where the traditional method's time consuming, multi-site process for flights, hotels, and tours, with stakeholders like agents and guides; land graphs features promise resolution.

Multi-agent coordination through a land graph streamlines vacation planning, with flight, hotel, car rental, and activity agents collaborating to book, update, and align arrangements.

Stateful systems remember your preferences, coordinate across agents, and adjust recommendations based on past choices to deliver personalized vacation planning.

Explore how cycles and persistence in land graph enable optimized vacation planning. Loop back to explore alternative dates, airlines, or routes, and pause and resume for flexible decisions.

Explore how Landgraf enables interactive and real-time features, including human-in-the-loop review and the ability to provide input at critical moments, plus streaming updates that support quick, informed decisions.

LangGraph draws inspiration from Tridgell, Apache Beam, and Network X to enable multi-agent communication, state management, and parallel data pipeline management within a unified, user-friendly interface.

Discover LangGraph world and how applications are organized as a graph to manage complex processes. Explain the three components—state, nodes, and edges—that govern task flow.

Explore how state acts as the memory of a food delivery tracking system, updating as tasks complete to reflect order details, status, delivery person, customer feedback, and payment status.

Define a Python state with a schema as the input for all nodes and edges, using a typed dict or Pydantic model and per-key reducers to control updates.

Explore node activation and flow control in land graphs, where each node performs a task and activates via edges, guided by supersteps and state graphs.

Understand how edges connect nodes to control a graph’s flow, with normal edges for linear progressions and conditional edges for state-based branching.

Explore the state graph, a flexible directed workflow that links nodes via a shared state, including a chatbot node, then compile and visualize it with mermaid.

Explore message graph as a state-based, message-centric model for conversational AI, using nodes like chatbot, start, end, and compile to manage dialogue flows.

Preview first graph in land graph to build a web-search chatbot that preserves state and messages, then add tools using Lang graph concepts and conditional edges to answer current events.

Build a chatbot using a Lang graph state machine and state graph, manage messages with add_messages, and employ a GPT-4 mini model via OpenAI, then test in a loop.

Enhance a chatbot by integrating tools like the Tavli search tool with the Lange chain framework, enabling real-time information retrieval and smarter responses.

Explore Rag, the retrieval augmented generation approach, to build real-time applications. Learn how retrieval, augmentation, and external information empower language models for accurate, contextually relevant responses.

Explore how a rag pipeline ingests private data, creates vector embeddings, indexes them in a vector store, and retrieves context for augmented generation using a large language model.

Explore the components of a retrieval augmented generation system, including the document loader, chunking, vector store, embeddings, retriever, and LLM, to enable accurate, relevant results.

Build a simple rag app by creating a retriever to load documents into a vector store, then assemble an agent with an agent node, a generate node, and a workflow.

Build a basic retriever by loading web documents, splitting into chunks, converting to vectors with OpenAI embeddings, storing in chroma, and enabling semantic retrieval for blog post queries.

The agent node acts as a decision maker, choosing built-in knowledge or vector store retrieval. It uses the retrieval tool and the generate tool to produce a contextually informed response.

Build the agent node by defining the agent state as a memory of messages, binding the retriever tool, and enabling GPT-4 to update the conversation history deterministically.

Define a generate function that uses the current conversation state and retrieved documents to produce a response with GPT four mini, zero temperature, streaming, updating the conversation.

Construct a Lang Graph workflow with an agent that decides when to retrieve using a retriever tool, then generate a final response, and compile the graph into a working system.