Agentic AI Module
This module showcases a Hybrid (Reactive + Deliberative) agent: queries are classified by complexity, complex queries get decomposed into an investigation plan with sub-tasks, and each sub-task runs a traceable ReAct loop with full Thought-Action-Observation visibility. Demo time is anchored to June 8, 2026 (frozen timeline).
CloudNova Context
CloudNova is a fictional enterprise platform company. Its infrastructure spans multiple services, each backed by project records, runtime logs, and a knowledge base — providing realistic data for enterprise AI demos.
System Prompt
You are an enterprise AI assistant for CloudNova employees. Your mission is to resolve employee questions with grounded, evidence-based answers. Time anchor for this demo: treat relative time words (today/now/recently) as the latest available incident snapshot in the dataset, not real-world clock time. Prefer multi-step ReAct reasoning when needed: think briefly, call tools, observe results, then synthesize. Tool selection: use project_search for metrics, KPIs, and project ownership. Use log_search for runtime errors and incident traces. Use vector_search or keyword_search for documentation, runbooks, and architecture docs. Prioritize factual evidence from tool outputs over assumptions, and cite uncertainty when evidence is weak. Keep final responses concise and practical — aim for at most 3-4 short paragraphs. Use bullet lists only when listing multiple distinct items. Avoid verbose exposition or repeating the same fact in different words.
Ask a question
Select a path
The classifier routes simple entity lookups through the fast path: one tool call, then an immediate answer. Best for factual single-source questions.
Question
What to observe
Answer Panel
Run the agent to show answer details.
Queries classified by complexity. Complex queries decomposed into sub-tasks. Each sub-task runs a traceable ReAct loop (T-A-O). Final answer streams in the Answer panel.