Generative AI: RAG, LangChain, GraphRAG & Fine-Tuning

Explore artificial intelligence, large language models, and APIs, and how prompts convert inputs into outputs like text, code, or summaries, revealing bias from training data.

Build a hands-on executive brief generator that converts meeting notes into a structured summary using llm prompts, guardrails, and a gradio user interface.

Learn how tokens drive cost, performance, and memory in LLMs, and how probability governs the most likely next token, with input and output tokens, prompts, and context limits.

Explore cost and latency basics in AI production, understanding tokens, prompt size, and model choice to optimize latency, cost, and quality.

Explore how temperature shapes randomness in llm outputs, balancing deterministic and creative results, and log prompts, tokens, latency, and cost in a lab.

Evaluate multiple models using the playground to compare accuracy, cost, and latency. Select the best model for your project based on the right answer and performance.

Explore a translation quality report project that compares multiple LLM translation models and uses an LLM as a judge to determine a winner by accuracy and latency.

Explore the output formatter project in module 3, learn to format output in a structured way, and review the GitHub source code with upcoming demo and implementation videos.

Transform unstructured data into a structured output format with fields like customer, issue type, product, and summary, enabling table-based analytics and insights.

Implement an output formatter that extracts structured JSON from user text, detailing customer name, issue type, product, priority, and summary, via a structured prompt and Gradio interface.

Learn to invoke an OpenAI model with LangChain, set up environment keys, and integrate tools like tabulate search for practical, prompt-driven LLM workflows.

Build a LangChain driven ai sql assistant that translates natural language queries into sql, validates and executes them with guardrails against SQLite database, and presents results via Gradio user interface.

Explore stateless llms, learn how they lack built-in memory and rely on sending current and historical messages as input tokens, shaping context limits and costs in long conversations.

Learn how LangChain memory automates storing and supplying conversation history for stateless LLMs, enabling context-aware dialogue with memory strategies like buffer, window, and summary memory.

demonstrates a customer support thread analyzer project, walking through the chatbot ui and memory strategies: stateless, buffer, window, summary, and token, plus a vector retrieval workflow with embeddings.

Master retrieval augmented generation (rag) and its steps, retrieval, augmentation, generation, for grounded answers from documents. See how chunking, embedding, and vector databases enable semantic search and private knowledge access.

Explore a simple rag demo using LangChain to load a pdf, split text into chunks, create embeddings with OpenAI, store vectors in Chroma, and answer questions via similarity search.

Build an on-call run book assistant that uses RAG with a Chroma DB vector store to search run books and provide instant, context-driven guidance.

Explore how GraphRAG enhances retrieval by modeling relationships between entities and connecting nodes beyond isolated chunks, contrasting graph databases with vector databases in retrieval augmented generation.

Conducts a hands-on demo of building and visualizing a knowledge graph with networkx and matplotlib, showing nodes and relationships, traversal, and a shortest-path reasoning from Alice to Falcon project.

Build a graph rag dependency explorer that combines graph traversal and vector search to map service dependencies and blast radius for on-call teams.

Explore the model context protocol (MCP) as a standardized interface that connects AI systems to tools, data, and external services, via tools, resources, server, and client roles.

Build an internal data assistant that converts plain English to safe read-only SQL, validates queries with guardrails, queries a SQLite database via MCP, and summarizes results.

Build a trading simulation agent that fetches market data from Yahoo Finance via YFinance and uses an LLM with guardrails to generate conservative buy, hold, or sell recommendations.