March 2026 — Pilot Complete

AI-Powered Tools for
Robot Engineering Teams

Four knowledge assistants, three web applications, and an automated welding system — built in 3 weeks to turn tribal knowledge into searchable, always-available tools.

19K
Lines of Code
7,800+
Messages Indexed
2,745
Documents Searchable
~50h
Development Time

Business Value

The Problem

Robot engineering teams accumulate critical knowledge through daily WhatsApp communication: weld parameter settings, error codes and fixes, procedure photos, assembly instructions, troubleshooting steps. This knowledge is:

  • Trapped in individual phone chat histories
  • Unsearchable — finding a specific fix from 6 months ago means scrolling through thousands of messages
  • Lost when people leave or change phones
  • Inaccessible to new team members who weren't in the original conversation

Telegram Knowledge Bots

C

Cloos Bot

Cloos robot programming team

4,326 WhatsApp messages from 8 channels → 1,763 indexed documents including 1,002 photo analyses

  • Search knowledge base with natural language or voice
  • Retrieve relevant photos and assembly videos
  • Save new notes back to the knowledge base
@Cloos_weld_bot
T

Teqram Bot

Teqram grinding robot service team

4,510 WhatsApp messages → 982 indexed documents including 676 photo analyses, 83 video analyses, and 14 technical documents

  • Search procedures, troubleshooting, and service history
  • Retrieve photos, videos, and technical documents
  • Covers April 2024 – March 2026 of team communication
@Teqram_grind_bot
Q

QC Bot

Quality Control inspectors

Project-based QC vault with intelligent photo categorization and automated report generation.

  • NDT inspection reports in three formats: Markdown, PDF, Excel
  • Batch report generation from Excel upload
  • Photo analysis and auto-filing by category (Welding / Coating / NDT)
  • WhatsApp bridge for field-to-office communication
@Quality_control_vault_bot
M

Main Bot

Internal / development

Central productivity assistant with 29 integrated tools spanning communication, file management, and code execution.

  • Google Workspace: email, calendar, Drive, Sheets
  • Voice input/output, photo analysis, image generation
  • Web search, code execution, persistent memory
  • Obsidian vault knowledge base integration
29 tools

Applications & Systems

F

Task Floor Planner

Iemants floor management

Web application for distributing tasks across the workplace floorplan with visual drag-and-drop editing.

  • Visual layout editor with the actual floor layout
  • 3-shift support with worker assignment per task
  • Project / phase / drawing metadata per task area
  • Photo and PDF attachments, priority flagging
Node.jsSQLiteVanilla JS
M

Cloos Maintenance App

Welding robot maintenance tracking

Maintenance management system with scheduled procedures, checklists, and full documentation support.

  • Dashboard with real-time alerts and task status
  • Procedure library with frequency scheduling
  • Operator checklists, measurements, and notes
  • Photo/PDF evidence, full-text search, 3-shift support
Node.jsSQLiteDocker~3,800 lines
Y

FPX — Multi-Layer Welding System

Yaskawa HC10DTP Cobot • Project 25380 Oranjewind

Automated multi-layer X-seam welding on 120mm thick steel plates, orchestrating 31 weld passes across both sides of the joint. Currently on-site testing.

  • Master orchestration alternating top (12 layers) and bottom (19 layers) for heat management
  • Automated ESAB weld job selection and validation via M-Registers
  • Per-layer position shifting, weave patterns, and arc control
  • Fault monitoring: wire / gas / water / general
  • Cloud VPS with ROS 2 Jazzy + MoveIt 2 prepared for monitoring dashboard
INFORM IIIYRC1000ESABROS 2
W

Weld Setting Visualizer

Interactive 3D welding physics

Real-time 3D simulator showing how welding parameters affect the weld — arc plasma, HAZ, grain structure, transfer modes. Single HTML file, runs in any browser.

  • 3D weld bead with thermal gradient and dime-stack normal map
  • Arc plasma with Fresnel glow — voltage controls height, current controls width
  • Transfer modes: short-circuit / globular / spray (EN transition currents)
  • Cross-section: banded HAZ zones with columnar dendrite grain structure
  • EN 1011-1 heat input → penetration depth and W:D ratio
  • Spatter physics, heat tint visualization
THREE.jsGLSL~3,000 lines

Measurable Results

Knowledge Recovery

7,800+ messages spanning April 2024 – March 2026, previously only accessible by scrolling phone screens, now searchable in seconds.

Photo Intelligence

1,761 photos analyzed with AI vision — each tagged with content description, defect types, NDT methods, and equipment. Searchable by description.

Report Automation

QC NDT reports that previously required manual Word/Excel work now generated from structured input in under a minute, in three formats.

Onboarding

New team members get instant access to years of institutional knowledge on day one — no more "ask Johan, he was there."

Technical Architecture

  Telegram Users (phone / desktop / web)
             |
             | Telegram Bot API
  +----------+----------+-----------+
  |          |          |           |
Main Bot  QC Bot   Cloos Bot  Teqram Bot
29 tools  9 tools   5 tools    7 tools
  |          |          |           |
ChromaDB  ChromaDB  ChromaDB   ChromaDB
(memory)  (QC vault)(Cloos KB) (Teqram KB)
  |          |          |           |
  +----------+----------+-----------+
             |
      +------+------+
      | AI Services  |
      | Claude       |
      | Gemini       |
      | Groq         |
      +--------------+

Technology Stack

LayerTechnologyPurpose
AI ReasoningClaude Sonnet 4.6 (Anthropic)Natural language understanding, tool orchestration
VisionGemini 3 Flash (Google)Photo analysis, content description
VoiceGroq Whisper Large V3Speech-to-text transcription
Vector DBChromaDBSemantic search over knowledge bases
EmbeddingsGemini Embedding 2 (768-dim)Document vectorization
Bot Frameworkpython-telegram-botTelegram integration, media handling
Web BackendNode.js / ExpressREST APIs, file handling
3D EngineTHREE.js r154 + GLSLWeld visualization, Fresnel shaders
Robot ControlINFORM III + ROS 2 JazzyYaskawa cobot welding programs
HostingHetzner Cloud (5 VPSes)Ubuntu 24.04, systemd, Docker

5 Production Servers

ServerSpecsRole
Main VPSCX43, 16GB RAMMain bot, Obsidian vault, central monitoring
QC Bot VPSCX23, 4GB RAMDedicated QC Bot
Cloos Bot VPSCX23, 4GB RAMCloos Bot + Teqram Bot
Website VPSCX23, 4GB RAMWeb apps, Cloudflare tunnel
Yaskawa VPSCX43, 16GB, 150GBIngestion center, future cobot control

Security & Operations

  • UFW firewall on all servers (SSH + required ports only)
  • fail2ban with SSH jail (brute-force protection)
  • Automated health checks every 5 minutes across all 5 servers
  • Auto-restart on service failure with Telegram alert notification
  • Daily automated backups (ChromaDB + vault data, 7-day retention)
  • Daily security summary reports

Knowledge Ingestion

1 Export
2 Parse threads
3 Photo analysis
4 Video analysis
5 Doc processing
6 Embedding
7 ChromaDB index
8 Hybrid search

Codebase

ComponentLines
Main Bot (bot.py + modules)4,521
Cloos Maintenance App3,795
Weld Visualizer (THREE.js + GLSL)2,983
QC Bot1,768
Teqram Bot1,223
FPX Welding System~1,200
Floorplan App1,135
Cloos Bot1,107
Ingestion Center + Monitoring~1,300
Total~19,000

Next Phase — Iemants Deployment

Deploy on Iemants Infrastructure

The platform is production-ready. This is a migration, not a rebuild — deployment can be completed in 1–2 working days once server access is provided.

Current (Pilot)Proposed
Personal Hetzner VPSesIemants server infrastructure
Developer-managedIT-managed with developer support
Individual bot per teamCentralized platform, team-specific knowledge bases
Web apps on external hostingServed from Iemants internal network
Personal API accountsCompany API accounts

What's Needed from Iemants IT

  • Server — Linux VM or dedicated server (Ubuntu 22.04/24.04, minimum 8GB RAM, 50GB disk)
  • Network — Outbound HTTPS to: api.anthropic.com, generativelanguage.googleapis.com, api.groq.com, api.telegram.org
  • API accounts — Anthropic (Claude), Google Cloud (Gemini), Groq (voice) — or continue with existing accounts on company billing
  • Telegram Bot tokens — Created via @BotFather on Telegram (free, takes 2 minutes)

How to Test

1

Install Telegram

Download from your app store (iOS / Android) or use the desktop app at telegram.org. Create an account with your phone number — free, takes 1 minute.

2

Get Your Telegram User ID

Search for @userinfobot in Telegram and send it any message. It will reply with your User ID (a number like 8594455361). Send this ID to Arne for authorization.

3

Start Chatting

Once authorized, search for the bot you want to test and send /start:

  • @Cloos_weld_bot — Cloos robot knowledge
  • @Teqram_grind_bot — Teqram grinding knowledge
  • @Quality_control_vault_bot — QC documentation & NDT reports

Example Questions to Try

BotTry asking…
Cloos Bot"What parameters for root welding on 8mm pipe?"
Cloos Bot"Show me photos of porosity defects"
Teqram Bot"What is procedure lin_20_circle?"
Teqram Bot"How to calibrate the grinding tool?"
QC Bot"Generate NDT report for project DAVIT CRANES"
QC BotSend a photo → bot analyzes and files it

Web Application Demos