LangChain4j brings AI capabilities to Java, which is ideal for building enterprise-grade systems that require transactions, observability, and security.
Agentic AI systems consist of a core LLM, tools, memory, and orchestration, with the key distinction being between programmatic workflows and autonomous agents.
Developing with local LLMs involves significant trial and error in model selection and prompt engineering, and requires handling issues like tool hallucination.
The workflow pattern uses programmatic code to orchestrate specialized agents in sequences, parallel tasks, or a mixture of experts for predictable outcomes.
Testing LLM-based systems requires new approaches like using sample-based evaluation, custom scoring functions, and strategies such as cosine distance or LLM-as-a-judge.
While workflows offer predictability and easier debugging, the agent pattern provides greater flexibility by allowing agents to autonomously decide which tools to use.
Agents can autonomously combine LLM capabilities with external tools like web services or search engines to accomplish complex, multi-step tasks.
Upcoming features include integration with the AITO protocol and a new programmatic API for composing complex agent interactions like sequences and loops.
msg
Ismaning, Germany
29:18Daniel Oh & Kevin Dubois
30:38Alex Soto
24:22Philipp Schmid
32:02Kevin Dubois
29:03Jörg Neumann
31:02Daniel Oh
39:11Thomas Dohmke
26:35Ricardo
Jobs that call for the skills explored in this talk.
Tata Consultancy Services Limited
Jacksonville, United States of America
Collaboration Betters The World GmbH
CommandLink
