VANTIQ Developers Guide – Introduction to Intelligence in VANTIQ Applications

What this guide is about. This developer guide explains the full spectrum of “intelligence” available to a Vantiq application and—more importantly—how to decide which kind to use where. Its central thesis is that real-time intelligent systems rarely succeed on a single technique; they compose four complementary forms of intelligence, each matched to a different kind of decision and its trade-offs in speed, certainty, explainability, adaptability, and autonomy.

  The four forms of intelligence it covers:

  – Knowledge-Based Rules — encoding what is known. Explicit domain expertise, policies, safety limits, and constraints expressed as deterministic conditional logic. Strengths are precision, low latency, auditability, and control; the limit is that rules only handle what is already anticipated and can suffer “rule explosion” as signal interactions grow. In Vantiq these are implemented with Visual Event Handlers (VEH) (the recommended low-code, DAG-based approach that generates VAIL behind the scenes) or directly in VAIL rules (WHEN EVENT OCCURS ON …), bound to a Service’s event interface or a Source.

  – Predictive AI — anticipating and detecting complex patterns and anomalies. Models trained on historical data surface weak signals, nonlinear correlations, and “shades-of-grey” anomalies that no explicit rule was written for. The guide grounds this with worked examples: an Azure ML Studio pump anomaly detector and a computer-vision pipeline scoring video frames via NVIDIA Triton over the Open Inference Protocol.

  – Generative AI — creating meaningful outputs from context and knowledge. Rather than selecting from predefined responses, generative models synthesize explanations, summaries, and recommendations from context and retrieved knowledge (including RAG, multimodal inputs, and personalization). Its role is interpretation and cognitive-load reduction—turning signals into meaning—illustrated by turning a raw pump anomaly into a grounded, plain-language mobile alert.

  – Agentic AI — goal-driven, outcome-oriented execution. Systems that adopt an objective and stay responsible for achieving it—planning, using tools, orchestrating services and people, and adapting as conditions change—in both single-agent and multi-agent (planner/executor/critic) architectures, with generative AI as the reasoning engine.

  The payoff — composing intelligence (“one size doesn’t fit all”). The guide closes by reframing the design question from “which form of intelligence is best?” to “what kind of decision is this, and what must it guarantee?” It maps decision types to architectural choices and presents a common composition pattern:

  1. Perception — rules catch known conditions instantly; predictive models flag anomalies and risks, emitting context-enriched, confidence-scored events.

  2. Understanding & Reasoning — generative AI synthesizes events, history, and domain knowledge into explanations and candidate actions.

  3. Execution & Adaptation — agentic AI owns the goal, orchestrates actions, monitors outcomes, and revises its plan as rules and predictive signals keep feeding it.

  The result is an integrated architecture that is constrained by policy (rules), informed by statistical insight (predictive), contextually understandable (generative), and capable of driving outcomes (agentic)—demonstrated in end-to-end scenarios such as a safety-and-security video application and a hospital-discharge application.

Click here to download a PDF of this guide