Building production AI systems — beyond the chatbot demo
Day 1
LLMs, APIs & RAG
- LLM fundamentals for developers: tokens, context windows, temperature, top-p, system prompts
- OpenAI, Anthropic, and open-source model APIs: practical comparison
- Structured output: JSON mode, function calling, tool use — reliable data extraction
- Prompt engineering for developers: few-shot examples, chain-of-thought, output formatting
- RAG (Retrieval-Augmented Generation): embedding models, vector databases, similarity search
- Vector databases: Chroma, Pinecone, pgvector — indexing and querying
- Chunking strategies: fixed-size, semantic, document-aware — impact on retrieval quality
- Reranking: improving retrieval precision with a cross-encoder
- Streaming responses: SSE and WebSocket delivery for chat UIs
- LLM integration in Spring Boot: Spring AI — chat, embedding, and RAG APIs
Day 2
Agents, Tool Calling & Production
- Agent architecture: ReAct pattern — reason, act, observe, repeat
- Tool calling: defining tools/functions, letting the LLM select and invoke them
- Agent frameworks: LangChain4j for Java, LangGraph for stateful multi-step agents
- Memory patterns: conversation history, summary memory, episodic memory
- Multi-agent systems: orchestrator-worker pattern, agent handoff
- Evaluation: LLM-as-a-judge, embedding similarity, RAGAS for RAG pipelines
- Guardrails: input/output validation, content filtering, hallucination detection
- Cost and latency: prompt caching, model routing, batching
- Observability for LLM apps: LangSmith, Langfuse, tracing LLM calls
- Production checklist: rate limits, fallbacks, user feedback loops
What your team walks away with
Developers who can build and ship production-grade AI features — from a simple LLM API call through a full RAG pipeline to a tool-calling agent with evaluation and observability.
- Call LLM APIs with structured outputs and tool definitions from a Java or Python backend
- Build a RAG pipeline with document ingestion, vector search, and reranking
- Implement a tool-calling agent using LangChain4j or Spring AI
- Evaluate AI feature quality with automated metrics and trace LLM calls in production
Book the AI + Agentic Development training
Hands-on throughout — participants build a working AI feature during the training. Can be extended to 2 days for deeper agent architecture coverage.
Get in touch