I've been in electrical construction for over 15 years. Started with installation, moved to commissioning, then became a chief project engineer. And you know what I realized over that time? The same mistakes happen over and over again. Missing tags, improper earthing, unsealed penetrations. Every single time. The same routine, eating up time and nerves.
📋 How I used to do inspections
First, I carried a notebook, a tape measure, and a camera. Later — a smartphone and Excel. Then came "IT-based technical supervision" — bulky, inconvenient systems that only work on a PC, require licenses, and are built for large corporations.
Even with them, inspections took hours. And worst of all — you had to configure them, train people, carry a laptop to the site.
I needed a tool that:
works in seconds,
requires no training,
is available right on the construction site,
and can be used by anyone — from a foreman to a technician.
💡 The idea
I'm not a programmer. But with the advancement of AI and neural networks, I realized I could make this happen.
Why not use a neural network to detect violations from photos?
I used:
YOLOv8 for image classification,
FAISS for similarity search,
Telegram as the interface — so I wouldn't need to build a separate app.
I collected a dataset: 500+ photos of violations and reference examples — from work, colleagues, and open sources. I trained the model for 50 epochs, then fine-tuned it on real-world photos.
The hardest part wasn't the neural network. It was keeping the bot stable on the server without running out of memory. It wasn't completely free — but the costs are minimal: cloud hosting and Render (~$7–25/month).
🤖 What the bot can do today
The bot now works in Telegram.
Send a photo of a panel, cable tray, or cabinet — and it instantly detects:
missing cable tags,
improper earthing,
unsealed penetrations,
untrimmed cable tray studs,
oxidation and rust,
missing single-line diagrams.
It generates a ready-to-use report with references to PUES, SP, GOST (for Russia) or IEC, NEC, BS 7671 (international). One photo — up to three defects. Free. No installation required.
It supports three languages: English, Russian, and Spanish.
🔥 What's next
I'm currently collecting feedback from engineers, foremen, and installers. I want to:
add more defect categories,
retrain the model on real photos,
export reports to PDF, integrate with project management systems.
🎯 Try it yourself
The bot is free. Just send a photo — get a report.
👉 @ElectricalInspectorBot
I'd love your feedback, ideas, and — of course — your photos of violations 😉