FOREMAN
MSIT AI-Track Capstone

FOREMAN

Production Intelligence with Memory

An AI platform that doesn’t just monitor industrial production environments — it remembers them. Four specialized reasoners on a biologically inspired memory substrate.

Open access

Try it yourself

A public demo instance is running — no registration, no request. The credentials below are deliberately shared so anyone can look around.

ACCESS

frontend-production-169a.up.railway.app ↗

E-mail
chef@foreman.de
Password
ForemanDemo2026!

The login is the plant-manager profile: it sees the whole fleet and may trigger the reasoners, so every capability is reachable from one account.

BEFORE YOU CLICK

Four honest notes

The interface is German. FOREMAN speaks the language of the shop floor it was designed for.

The data is simulated. No real plant is connected. Every prediction carries a visible simulation caveat — a deliberate design rule, not a placeholder.

It is a shared instance. Anything you enter is visible to everyone else and stays in the database. Please do not enter real personal data.

Nothing here switches anything. FOREMAN explains, it does not actuate. There is no path from this interface to a machine.

The pitch

FOREMAN in ten slides

The full project deck — problem, three-layer architecture, its reasoning layer, and how it is built. Interactive; use the arrows to navigate. Open in full screen ↗

FOREMAN · deck
Architecture

Three layers, cleanly decoupled

Industry delivers the data, FOREMAN reasons, operators act. Each layer is autonomous — the platform is an open node in the industrial software stack, not a closed app.

LAYER 1

Industrial environment

Sensor streams, PLC logic (OPC UA, MQTT, Modbus), log files, and maintenance history — integrated through normalizing, anonymizing adapters.

LAYER 2

Reasoning platform

Four specialized reasoners work against a shared memory substrate, routed through a local + cloud model gateway.

LAYER 3

Output channels

An operator dashboard with traceable explanations — plus an MCP interface that any third-party system can consume.

Layer two

Four reasoners, one memory core

Each reasoner answers a question that a snapshot dashboard cannot.

01

Event-chain reconstruction

Links incidents, sensor drift, and operator logs into coherent narratives over time.

02

Drift detection

Flags changes in a machine’s behavior against its own historical profile.

03

Failure prediction

Matches the current state against consolidated precursor patterns from past incidents.

04

Maintenance-cycle analysis

Causal deviation analysis between maintenance events and machine behavior to optimize intervals.

Load data, not load simulation. FOREMAN does not simulate load itself — that needs parameters outside its observation boundary. It exposes the observed load profiles and limits read-only over MCP, for an external simulation tool to build on.

Engineering standards

Built to be reviewed, not vibe-coded

Every change passes defined gates before it reaches main. The binding definition lives in the repository’s GROUND_TRUTH §10–§11.

Type-safe mypy --strict / tsc
Tested pytest ≥ 85% coverage
Linted ruff / eslint
Secure OWASP Web & LLM Top 10 (2025)
Private by design GDPR Art. 25
EU AI Act classified before code
Observable per-reasoner metrics (OWASP A09)
Human-in-the-loop operator confirms safety calls (BSI)
Bounded rate-limited + pinned models (LLM10)
Living docs updated in the same commit