Skip to main content
Back to Projects

Industrial Diagnostics

Industrial Edge Discovery

Diagnostics and Live System Documentation

A read-only platform for industrial edge support teams. It connects to remote devices, discovers running services and dependencies, generates live documentation, and shortens root-cause analysis with snapshot-based diagnostics.

Read-only by default
Brownfield + Greenfield
Deterministic discovery first

4

Architecture Layers

2

Operating Modes

6

Diagnostic Steps

Preview

Device Snapshot

Last scan 2m ago

edge-berlin-03

Ubuntu 22.04 · uptime 68d · VPN + SSH

opcua-collector

Healthy · last ingest 34s ago

mqtt-bridge

Warning · broker handshake failed

timescaledb

Healthy · last write 3m ago

Likely fault domain

Upstream broker connectivity after recent endpoint change

Snapshot diff found a new MQTT endpoint and no collector-side regressions.

01

Problem

Industrial edge deployments drift over time. Containers get added, configs change, documentation falls behind, and support engineers often begin incidents without a reliable picture of what is actually running on the target device. Troubleshooting becomes slow, expert-dependent, and inconsistent across customers.

Support Starts Blind

Incidents begin without a trustworthy view of services, ports, logs, or recent changes on the target device.

Documentation Drifts

Real deployments evolve after go-live, but drawings and handover docs rarely keep pace with brownfield changes.

Expert Bottlenecks

Troubleshooting depends on whichever engineer remembers the customer-specific setup from six months ago.

02

System Flow

The core value is not a chatbot. The platform works by securely collecting technical facts, building a device model, and then using that model to generate documentation and guided diagnostics.

Access Layer

Remote Edge Device

Approved VPN, SSH, or customer remote-access tooling establishes read-only access.

Docker containers

System services

Logs and configs

Databases and ports

Knowledge Model

Inspect, classify, understand

Discovery Engine

Normalizes raw inspection data from the device

Role Classifier

Identifies collectors, dashboards, databases, and unknowns

Snapshot Diff

Compares current state against previous and known-good baselines

Live Dossier

Auto-generated documentation with service inventory, topology, and storage paths.

Health Summary

Detects stale collectors, stopped services, and disconnected brokers.

Guided Diagnostics

Highlights likely fault domains and the next logs or configs to inspect.

Step 1

Connect securely

Step 2

Inspect system state

Step 3

Classify components

Step 4

Generate dossier

Step 5

Compare snapshots

Step 6

Guide diagnostics

Brownfield Mode

Connects into messy, already-running customer systems and infers structure from what is actually deployed.

Useful immediately on existing fleets

Handles incomplete standards and unknown services

Ideal for audits, support, and handovers

Greenfield Mode

Uses naming conventions, labels, and health endpoints from standardized deployments to improve precision.

Higher-confidence classification

Cleaner documentation and automation

Better long-term change tracking

03

Technical Approach

Designed the platform around deterministic discovery rather than a chat-first UX. A secure access layer connects over approved remote methods, the discovery engine inventories Docker, services, logs, configs, databases, and network state, and a knowledge model turns raw inspection into a structured system view. Versioned snapshots enable drift detection and guided diagnostics, while AI remains an optional explanation layer on top of a rules-driven foundation.

snapshot/device-model.json
{
  "device": "edge-berlin-03",
  "services": [
    { "name": "opcua-collector", "role": "collector", "status": "healthy" },
    { "name": "timescaledb", "role": "database", "status": "healthy" },
    { "name": "mqtt-bridge", "role": "bridge", "status": "warning" }
  ],
  "drift": [
    "mqtt endpoint changed",
    "new collector-debug container"
  ],
  "diagnostics": {
    "faultDomain": "upstream connectivity",
    "nextChecks": ["mqtt-bridge logs", "broker endpoint config"]
  }
}

Design Principle

Reliable inspection comes from evidence first. AI is only useful after the platform has mapped the real system state.

Deterministic discovery first, AI explanation second.

Read-only by default

No automatic config changes

Auditable checks and evidence trails

Customer-controlled credentials

04

Platform Modules

Module 1

Connector Module

Module 2

Discovery Module

Module 3

Parser and Classifier

Module 4

Knowledge Model

Module 5

Snapshot and Diff

Module 6

Diagnostics Engine

Module 7

Report Generator

Module 8

Optional AI Assistant

05

MVP Scope and Outcomes

Recommended MVP

Connect to one industrial edge device and generate a useful system report plus basic diagnostics.

SSH-based access to the target device

Docker, service, config, log, and database discovery

Basic role classification and health summary

Versioned snapshots and snapshot comparison

Generated technical dossier for support handover

Outcome 1

Defined a four-layer architecture from secure access through diagnostics

Outcome 2

Scoped a read-only MVP around discovery, generated dossiers, and snapshot comparison

Outcome 3

Modeled brownfield and greenfield operation for messy and standardized fleets

Outcome 4

Separated deterministic discovery from optional AI explanations for industrial trust

Outcome 5

Designed drift detection against previous and known-good snapshots

06

Tech Stack

SSH
Docker Discovery
systemd
Snapshot Diffing
Rule-Based Diagnostics
Knowledge Graph Modeling
TypeScript
Next.js