How I used QlikSense Enterprise as an AI orchestration layer — ETL, API trigger, visualization and pseudo real-time alerting

[Mhena]

Hi QlikSense Community,

I'd like to share an architecture where QlikSense Enterprise on Windows plays a central role across four layers of an AI pipeline — not just visualization, but also data preparation, API orchestration, and pseudo real-time alerting via NPrinting.

Context

The project runs on a manufacturing production line. The goal is to detect machine drift automatically, before it impacts quality. The AI model is a LSTM autoencoder — it learns the normal behavior of the machine and flags anomalies when the reconstruction error exceeds a threshold. The model is deployed as a REST API on private infrastructure.

4 roles for QlikSense Enterprise

1. ETL — Data preparation

QlikSense loads raw inspection files from a network folder, applies transformations and aggregations through its scripting engine, and stores results as QVD files. A listener mechanism at each reload detects whether incoming data has already been processed. If nothing is new, the reload is intentionally stopped so the app does not refresh unnecessarily — keeping the pipeline efficient and avoiding redundant API calls.

2. API trigger — Inference orchestration

QlikSense calls the LSTM autoencoder API via a native REST connection. It sends a sliding window of the most recent measurement points as input to the model. The call sequence is carefully designed so that only one request triggers inference — subsequent calls read results from the API without re-running the model, preserving the integrity of the AI state.

3. Visualization — Drift dashboard

Results returned by the API are stored as QVD files and loaded into a visualization app. Charts show the reconstruction error per feature over time, with detected drift sections highlighted. Timestamps are displayed in local timezone.

4. Pseudo real-time alerting — NPrinting

When drift is detected, an alert is automatically distributed via NPrinting. The reload cycle acts as the pseudo real-time trigger — each time a new file arrives from the machine, the pipeline runs end to end, and if the model detects a drift, NPrinting generates and distributes the report to the relevant recipients immediately. No manual intervention required.

Historical replay

To backfill historical data, a dedicated script loops over a historical QVD with a sliding window, calls the API for each window, and accumulates results progressively. The request rate is controlled to stay within API limits.

Key takeaway

QlikSense Enterprise is much more than a BI tool. With REST connections, scripting logic, QVD persistence, and NPrinting distribution, it can act as a complete lightweight orchestration layer for an AI pipeline — without any additional middleware, and within an on-premise environment.

Happy to answer questions or share more details on any of the four layers.