HAIIP

Human-Aligned Industrial Intelligence Platform

Production-grade AI for Nordic industrial SMEs — predictive maintenance, anomaly detection, EU AI Act compliant

RDI deliverable

---

The Problem

Nordic industrial SMEs lose EUR 500–2000 per hour of unplanned downtime. They cannot afford data scientists. They cannot deploy black-box AI that operators distrust. From August 2026, every AI system must comply with the EU AI Act.

HAIIP solves all three.

---

What HAIIP Does

Capability	How
Real-time monitoring	OPC UA, MQTT, vibration CSV, built-in simulator
Failure prediction	Gradient Boosting + BiLSTM, explains why in plain language
Human-in-the-loop	Operator can override every AI decision
Closed-loop control	Decision reaches the machine actuator, not just a dashboard
EU AI Act compliance	Full audit trail, model cards, drift detection
RDI reporting	Generates evidence artifacts for EU-funded projects

---

Architecture

Data Sources          AI Core                     Interfaces
------------          -------                     ----------
OPC UA (PLC)    -->   AnomalyDetector (IF)        Streamlit HMI (10 pages)
MQTT broker     -->   MaintenancePredictor (GB)   FastAPI REST + WebSocket
Vibration CSV   -->   AnomalyAutoencoder (LSTM)   /api/docs (OpenAPI)
Simulator       -->   MaintenanceLSTM (BiLSTM)    Demo mode (no auth)
                      DriftDetector
                      AutoRetrainPipeline
                      RAGEngine + LLM Agent
                      ComplianceEngine

Stack: Python 3.11 · FastAPI · Streamlit · scikit-learn · PyTorch · asyncpg · Redis · Docker Compose

---

Quick Start

git clone https://github.com/Aliipou/HAIIP.git
cd HAIIP
docker compose up --build

• Dashboard: http://localhost:8501
• API docs: http://localhost:8000/api/docs
• Demo mode works without real hardware

---

Project Structure

HAIIP/
├── src/
│   ├── ai/              # Models: anomaly detection, predictive maintenance
│   ├── compliance/      # EU AI Act compliance engine
│   ├── data/            # OPC UA, MQTT, CSV ingestion
│   └── api/             # FastAPI application
├── ui/                  # Streamlit HMI
├── docs/
│   ├── MODEL_CARD.md
│   └── DATASET_CARD.md
└── docker-compose.yml

---

EU AI Act Compliance

HAIIP targets Limited Risk classification under Article 52:

• Full audit trail for every AI decision
• Model cards and dataset cards
• Drift detection with automatic retraining
• Human override at every decision point
• Explainability layer (SHAP + plain-language summaries)

---

License

MIT