Ah, got it! Here is the updated translation with “SheeGurBot”:
I realized we needed a “Waze for rocket alerts” – so I built SheeGurBot. 🚀🚨
Living in Israel means being plugged directly into an endless stream of information sources.
Telegram channels, WhatsApp groups, alert apps – the critical info is always there, but it’s buried under mountains of noise: repetitive messages, ads, technical updates, and minor alerts that flood your phone exactly during the moments when you need quiet and focus.
The motivation behind my latest mini-project was simple: Maximum clarity, minimum noise, minimum delay.
I wanted a system that would monitor all the sources I trust, understand exactly what’s happening using Artificial Intelligence, and send only the most relevant, distilled information straight to my WhatsApp.
What happens “under the hood”?
It’s not just a bot that “forwards messages.” To achieve a high level of reliability, I built an engine in Node.js (meet SheeGurBot) that manages several complex layers:
🔹 Real-time multi-source monitoring – The bot listens simultaneously to Telegram channels (via gramjs) and WhatsApp groups/chats (via whatsapp-web.js).
🔹 AI-powered classification – Every message goes through an LLM-based classification layer. Today, it’s no longer a single model, but a fallback system between several providers and models based on availability/blocks/errors, including EdenAI and Google AI APIs.
🔹 Understanding meaning, not just text – The model doesn’t just read the message; it classifies the event type (launch, interception, impact, return to routine, supplementary info, etc.), identifies relevant areas, assesses urgency, and cleans out ads, links, and textual noise.
🔹 Global deduplication – If five sources report the same event, the bot identifies the overlap (using AI + fuzzy matching + the history of messages already processed by the system) and sends only the clearest, most relevant version.
🔹 Full personalization – I developed a command-based registration system. Every user can define exactly which types of alerts interest them, the level of detail they want, and in which geographical areas (Gush Dan? The Galilee? The confrontation line?). The system filters out the white noise and sends each person only what’s critical for them.
🔹 Smart queue management – The system knows how to prioritize. Urgent messages are sent immediately, less urgent updates can “wait” a bit to be merged, and if something urgent is already going out – the bot knows how to attach what’s already waiting in the queue to it, so it doesn’t make your phone ping 10 times a minute.
The Result
Instead of a flood of messages, you get a clean and accurate stream:
🤖 🚀🚨
📍 איזור: השרון
⏱ זמן הגעה משוער: 20:28:21
------
🕐 זמן זיהוי: 20:22:21
השיגורבוט
This was a fascinating exercise in combining real-time data processing, queue logic, deduplication, personalized filtering, and the reasoning capabilities of LLMs — to solve a problem that affects the daily lives of us all. The AI I used learned a lot from this.
And finally, after the AI finished sorting out all the debugging, fallbacks, and prompt edge-cases with me, I asked it to help me write the story of the process for this post as well 🤖✨
For anyone interested in the SheeGurBot:
#BuildInPublic #NodeJS #GenerativeAI #LLM #SoftwareEngineering #RealTimeSystems #IsraelTech #AI #Automation #Personalization
Leave a Reply