Python frameworks for production LLM applications
Module 1
LangChain — Chains, Agents & Tools
- LangChain architecture: LCEL (LangChain Expression Language), runnables, chains
- LLM and Chat Model wrappers: OpenAI, Anthropic, Ollama (local models)
- Prompt templates: ChatPromptTemplate, few-shot templates, partial templates
- Output parsers: PydanticOutputParser, JsonOutputParser, StrOutputParser
- Memory: ConversationBufferMemory, summary memory, entity memory
- Tool calling: defining tools with @tool decorator, ToolNode
- ReAct agent: LangGraph-based agent with tool use loop
- LangGraph: stateful multi-step agents — state machine for LLM workflows
- Retrieval chain: RetrievalQA, RAG chain with LCEL
- LangSmith: tracing, evaluation, and prompt versioning
Module 2
LlamaIndex — RAG Pipelines & Advanced Retrieval
- LlamaIndex architecture: documents, nodes, indices, query engines
- Document loaders: PDF, HTML, Notion, database, custom loaders
- Node parsers: chunking strategies — sentence, semantic, hierarchical
- Embedding models: OpenAI, Cohere, local (sentence-transformers)
- Vector stores: Chroma, Pinecone, Weaviate, pgvector integration
- Index types: VectorStoreIndex, SummaryIndex, KnowledgeGraphIndex
- Query engines: RetrieverQueryEngine, router query engine
- Reranking: Cohere Rerank, cross-encoder reranking
- Sub-question query engine: decomposing complex questions
- Evaluation: RAGAS — faithfulness, answer relevancy, context precision
- LlamaIndex vs LangChain: when to choose which or combine both
What your team walks away with
Python developers who can build RAG pipelines and tool-calling agents with LangChain and LlamaIndex — with proper evaluation and LangSmith observability.
- Build a RAG pipeline with LlamaIndex from document ingestion through reranked retrieval
- Implement a LangGraph tool-calling agent with memory and multi-step reasoning
- Evaluate RAG pipeline quality with RAGAS metrics
- Trace LLM application calls with LangSmith for debugging and improvement
Book the LangChain & LlamaIndex training
A focused hands-on day — participants build a working RAG application. Works best after the Prompt Engineering or AI + Agentic Development training.
Get in touch