Detecting behavioral pattern shifts across your crypto Telegram groups before price moves happen

If you're in a lot of crypto Telegram groups you've probably noticed that the tone and activity in those groups often shifts before big moves happen. The problem is when you're in hundreds of groups you cant read every message so you miss the signals until after the move already happened.

I built something that tracks this automatically across all your groups and uses on-device NLP to detect 9 different pattern types. Heres what it actually looks for:

Sentiment shift detection with Soros-style reflexivity analysis. It tracks sentiment velocity and acceleration across groups over time. When sentiment is shifting fast AND accelerating in the same direction thats reflexive momentum and it gets flagged as critical severity. If 15 of your groups all flip from bullish to bearish within the same window thats a very different signal than one group having a bad day.

Market psychology signals that detect fear capitulation and overconfidence breakdowns. It builds an uncertainty index and confidence index per group over multiple days and watches for flips. High uncertainty suddenly dropping to high confidence = potential bottom signal. High confidence suddenly collapsing into uncertainty = potential top signal. It measures the flip strength using gradient and acceleration so it can distinguish a slow drift from a sharp reversal.

Address propagation tracking. When the same wallet address or contract address starts appearing across multiple unrelated groups simultaneously thats either coordinated shilling or organic discovery of something new. Either way its useful to know about before everyone else notices. It detects Ethereum, Solana, Bitcoin and other chain addresses automatically.

Topic propagation. When the same token name or topic starts spreading across groups that don't normally discuss similar things it flags the cross-group spread with a timeline showing which groups picked it up first and how fast it moved.

Silence anomaly detection. When a group thats normally active suddenly goes quiet that can mean insiders know something the rest of the group doesn't or that a rug is about to happen. It compares recent message volume against historical baselines and flags statistically significant drops.

Urgency clustering. When multiple groups simultaneously spike in urgent language like "buy now" "last chance" "don't miss this" it detects the cluster and shows you which groups are affected and whether it looks coordinated or organic.

Question clustering. When people across different groups start asking the same questions at the same time like "is X a scam" or "why is Y dumping" it surfaces the pattern before you would have noticed it from scrolling.

Entity mention surges. Tracks when a specific token project or person suddenly gets mentioned way more than their baseline across your groups.

It also has real-time smart alerts where you set up rules like "alert me when anyone mentions $TICKER in these groups" and it evaluates every incoming message in real time against your rules. So if you're tracking a specific token across 50 groups you get notified immediately when it comes up anywhere.

On the social graph side it tracks which users are cross-posting the same content across multiple groups and builds profiles showing posting patterns group overlap and propagation behavior. Useful for identifying coordinated pump campaigns or figuring out which accounts are signal sources vs just noise.

Group discovery is also built in. Telegram locks its similar channels feature behind Premium and caps results. This scans all your groups and cross references similarity data into a ranked list of hundreds of new groups. It handles the FLOOD_WAIT rate limiting automatically if you want to bulk join so you dont get locked out after 10 to 15 joins.

Everything runs locally on your phone. No servers, no cloud processing, no data leaves your device. Completely free, no subscription, no premium tier.

Happy to answer questions if anyone wants to try it.