LLM chatbot design - You will build chatbots powered by local and cloud LLMs, configuring conversation memory, context handling, and custom response handlers for practical business scenarios.
End-to-end RAG pipeline - You will learn how to assemble a full RAG pipeline, from source documents through embeddings and a vector store to answer generation with relevance checks and quality control.
Document and media processing - You will master data preparation for PDF, DOCX, HTML, TXT, and multimodal sources, including OCR and video captions, so your AI systems can use clean and structured knowledge.
Vector database selection - You will learn how to choose a local or cloud vector database, build indexes, filter metadata, and update resources so semantic search stays accurate, stable, and efficient.
Better prompts and routing - You will create dynamic prompts, conditional chains, and query routing across models, databases, and knowledge sources, making responses more relevant and useful to users.
AI workflow automation - You will design workflows built with chains, agents, and classifiers that ask follow-up questions, detect user intent, and guide people through the next steps automatically.
Error diagnosis and control - You will learn practical ways to detect empty retrievals, weak answers, and hallucinations, then add fallbacks, validation, and monitoring to improve reliability after launch.
Secure solution deployment - You will be prepared to deploy AI systems with access control, data protection, GDPR awareness, action traceability, and a solid approach to ongoing maintenance and expansion.
LangChain provides a modular foundation for structuring AI applications instead of treating model calls as isolated steps. This section maps the roles of models, chains, agents, and tools within a workflow. Typical use cases include chatbots, RAG systems, recommendation scenarios, and task automation.
Local and cloud-hosted LLMs introduce different integration patterns, constraints, and control points. The section covers connecting local models through REST API or libraries, as well as configuring API keys and limits for remote services. It also includes conversational memory, context handling, and custom query handlers.
RAG changes the way language models answer questions by retrieving external knowledge before generation. This section contrasts that approach with a classic standalone LLM workflow and explains the retrieval-to-response sequence. It also positions LangChain as a framework for building flexible RAG pipelines.
Before retrieval can work well, source materials must be normalized and divided in a usable way. The section covers handling PDF, DOCX, HTML, and TXT files, along with segmentation and metadata extraction. It also includes initial content cleaning and chunking decisions that shape later stages.
Choosing a vector database affects retrieval strategy, maintenance model, and deployment setup. This part compares local and cloud options, then moves to index creation and update workflows. It also addresses metadata filtering and personalization mechanisms that refine search results for different queries.
Embeddings determine how content is represented for later semantic retrieval. Participants compare available embedding models, then review how to create and test custom embeddings for specific needs. The section also explains how these representations are connected to the larger RAG pipeline.
A lightweight classifier can improve decision-making before retrieval or generation begins. The section explains when such a classifier is useful, how to choose features and labels, and how to define the prediction structure. It then shows how classification fits into a broader decision pipeline.
From source retrieval to final answer, a RAG pipeline must be assembled as one coherent flow. This section connects embeddings, database operations, and response generation, then examines how to test relevance and answer quality. It also covers diagnosing empty results, poor responses, and hallucinations.
MCP in LangChain is presented through the path data takes from user input to generated output. Participants analyze the roles of input, processing, and generation components, and how they interact inside a workflow. Practical examples help relate the architecture to concrete implementation patterns.
Building with LangChain means shaping how processing steps connect and react to input. This section discusses simple, sequential, and parallel chains, along with handling dynamic parameters. It also addresses data transformations and conditional flow control so outputs can influence the next stage of execution.
Dynamic prompting links model instructions with application logic and available context. The section explains the difference between prompts used as fixed templates and prompts assembled from rules. It also covers variables, historical and dependent data, plus adaptive behaviors when required information is missing.
Agents can move a dialogue forward instead of waiting for perfectly formed input. In this section, the focus is on follow-up questions, intent recognition, and query classification that support automated interaction. Context-aware conversation with memory is used to keep exchanges consistent across turns.
Combining several knowledge sources in one workflow requires clear routing and context control. Participants examine how RAG can work with MCP and additional repositories, and how queries are directed to different models or databases after classification. The section also covers switching between contexts and sources.
Conversation quality often breaks down when context is lost between turns or sessions. This part addresses session handling, storage and reuse of chat history, and methods for managing incomplete user input. Fallback strategies are discussed as a way to keep flows usable despite missing data.
Vector knowledge bases rely on careful preparation before retrieval quality can be improved. Here the focus is on cleaning data, segmentation, metadata design, and choosing an appropriate chunking strategy. The section also includes index creation, versioning, updates, semantic search, and metadata filtering.
Working with real knowledge sources requires more than plain text ingestion. The section covers extraction from PDF, office files, HTML, and other structured or unstructured content, then extends to OCR, images, and video subtitles. It also shows how multimodal inputs can feed one RAG flow.
As AI usage grows, individual workflows must be turned into repeatable and scalable processes. The section addresses how to organize execution so solutions handle higher load, recurring tasks, and broader operational use. Attention is given to maintaining control over reliability as automation expands.
Security in AI systems depends on both technical safeguards and accountable operation. Participants work through data storage, access control, response validation, and ways to reduce hallucinations. The section also addresses GDPR, privacy, compliance, and auditability of system decisions.
Once a solution is in production, logs become the main source for improving it. This section focuses on registering user queries, reviewing conversation traces, and tracking response quality with a feedback loop. It also covers refactoring and extending existing chains and prompting logic.
Programming basics - You should be comfortable reading and editing simple code, running scripts, and understanding variables, functions, and basic data structures used in AI integrations.
API and JSON basics - You should understand how API communication works, be able to read requests and JSON responses, and know what API keys and basic authorization are used for.
Working with data - You should have basic experience with text files and documents and understand that data often needs cleaning, segmentation, and preparation before it is used in an AI system.
LLM fundamentals - You should know the general idea behind language models, understand what prompts and context are, and be aware that a model can produce incomplete or incorrect answers.
