NUILab Agentic AI
NUILab Agentic AI
Patterns
Reusable design patterns for agentic systems — orchestration, tool use, planning, memory, RAG, multi-agent, and evaluation.
Tool use
The pattern every agent is built on: give the model a set of tools it can call, and loop — it picks a tool, you run it, the result goes back in, and it goes again.
PatternsPlanning
Have the agent break a goal into steps before it acts, work the plan, and revise it as it learns — so a multi-step task doesn't wander or lose the thread.
PatternsMemory
What an agent carries between steps and across sessions — the running context, notes it writes down, and facts it pulls from a store — and how to manage a context window that is always too small.
PatternsRetrieval (RAG)
Fetch the documents relevant to a question and put them in the model's context before it answers, so it works from your actual data and can cite sources instead of guessing from training.
PatternsEvaluation
How to tell whether an agent's output is any good — systematically, with deterministic checks, judge agents, and a fixed test set — instead of eyeballing one run and hoping.
PatternsMulti-agent orchestration
Run several scoped agents instead of one that does everything — for parallelism and for focused context — and coordinate them with a manager, a pipeline, or a shared bus.