Crypto trading tools with our own ML. Node-based strategy builder, AI-annotated charts, NN entry/exit detection. News & fundamentals built in. Beta soon
Hey everyone! This is the Caeris Lab team.
Today we're officially launching our pages on X, Telegram, and Reddit. This is where you'll get first-hand updates on our development, team ideas and insights, polls on new features, and much more.
What are we building?A unique trader's workspace that brings everything you need into a single place. We're currently around 40% through the planned functionality.
What's inside:
- Levels, indicators, and patterns
- News feed with built-in AI analysis
- A powerful neural network - your trading companion that automatically scouts entry and exit points
- A strategy builder powered by all our tools (AI, news, and more)
- Correlation of current market conditions with historical data
- And much more...
Below you'll find screenshots of our strategy builder and the data correlation interface. Heads up: this isn't the final version - the UI will keep evolving.
Stay tuned - the best is yet to come π
#AITrading #AlgoTrading #Fintech
70.5% win rate π― 1.32 profit factor π₯
Continuing work on Caeris Lab AI Phase 1!
We managed to seriously boost the macro F1 and balance out the classes, so now all 5 classes are being used pretty much evenly
On top of that - even though it's still early - we decided to add a few metrics like drawdown, PNL , win rate, profit factor, and more. As you can see in the screenshot, the results are pretty solid π, especially considering this is just one piece of a much bigger NN
We also ended up picking 5 main timeframes: 1m, 5m, 15m, 30m, 1h. That doesn't mean trades will be opened on exactly these intervals - it just means this specific part predicts the price direction across these 5 timeframes (in the chart you see h1 instead of 1m because that's horizon 1)
So while part of the team is working on our NN, the other part is working on the site. And we've decided to introduce a Leaderboard system (kind of similar to the one on Binance)
The idea behind the table is that users who don't want to (or can't) - build a solid strategy with our builder can just copy the trades of the top leaders by PNL/Score, and that way they end up making the same trades the leaders do.
Right now each leader has a little summary of their trades for easier analysis (the numbers are made up for now and don't reflect any real stats).
The thing is, we still haven't figured out what kind of reward to come up with for the leaders whose trades get copied. A good solution would be a % of the commissions, but since users will be trading with money that's already on the exchanges (our service doesn't hold their funds), we can't deduct any commissions
So we've been working on Phase #1 of Caeris Lab AI for 4 days now, and here are a few interesting takeaways:
- We managed to improve the model's accuracy by almost 20% compared to last time
- Instead of 5 classes at the output of each horizon, we tried a logit distribution to better catch the signal. It did boost the information content of the features (IC), but paradoxically it dropped the model's accuracy - meaning the model was catching a signal, just not the right one
- We built a new NN architecture based on the Temporal Fusion Transformer, but unlike the original, our model is way better at telling the different horizons apart and doesn't try to let one horizon override another
Ideally we'd hit 40% accuracy for Phase 1 - that would give us a solid starting point. But as the tests on smaller timeframes show, more often than not you get noise rather than a real signal
#algotrading #crypto
News about Caeris Lab AI π΅
When the model screams "strong up" it's right 35% of the time.
When it calls "strong down", 39%. Random would give you 10%.
That's almost a 4x edge on the strongest signals.
1-hour horizon, BTC and SOL.
Eleven iterations deep at this point, each one moving the needle.
Latest update: hooked up our production news-analysis engine straight into the model. It chewed through 106k crypto articles in 2 hours and tagged each one with which coin it's actually about (not just casually mentioned) and whether the tone is bullish or bearish on that coin.
Now the model can tell the difference between "SEC sues Coinbase"
and "XRP got name-dropped in an altcoin recap".
Next up: per-coin models (Bitcoin and Solana split out separately)
and moving from "up/down" calls to actual magnitude forecasts.
We're kicking off development of Caeris Lab AI! π
It's going to be made up of several NN that interact with each other (like agents)
Phase #1 - Multi Horizon NN
The idea is that it should predict one of 5 price movement scenarios:
- strong down
- down
- flat
- up
- strong up
across 5 different timeframes (we haven't locked down exactly which ones yet - that'll come out of testing)
Just like with News AI, we'll be sending out reports on how development is going as we go.
Fingers crossed for great results! π€
Weβve been discussing with the team the potential of adding a Twitter (X) AI - a model that would track the most influential accounts, pull their posts in real time, and analyze them. Do you think it makes sense?
Added Lookback to the strategy builder. We decided to test a very simple strategy right away.
Parameters:
- Leverage: 1x
- Starting balance: $18K
- Position size: $1K
- Fee (open/close): 0.05% (0.1% total per trade)
- Range: 01-06-2026 β 05-26-2026
Results:
- Trades: 28
- Profit: +$6,112.99 (after fees)
- In %: +33.96%
- Max drawdown (open position): 17.22%
- Average trade duration: 4.3 days
- Average drawdown (open position): 1.54%
Like I said, this was an extremely simple strategy - no stop-loss, no take-profit, no additional signal confirmation logic - and it still generated profit.
Just imagine how powerful itβs going to be at launch π
π’ Whitelist is officially live! π
https://t.co/OLHaEaHu9v
We've launched the starter version of the site! Right now it has a couple of sections we think you'll find interesting. We also added a roadmap π in the form of a star map - definitely check it out!
Prices aren't listed yet because we're still in the whitelist stage. They'll be added once the product officially launches.
Some whitelist users, chosen at random, are gonna get some nice gifts! π
And don't forget about our Telegram group - there's gonna be a lot of cool stuff there soon! https://t.co/2Il8jivyNM
π’ Whitelist is officially live! π
https://t.co/OLHaEaHu9v
We've launched the starter version of the site! Right now it has a couple of sections we think you'll find interesting. We also added a roadmap π in the form of a star map - definitely check it out!
Prices aren't listed yet because we're still in the whitelist stage. They'll be added once the product officially launches.
Some whitelist users, chosen at random, are gonna get some nice gifts! π
And don't forget about our Telegram group - there's gonna be a lot of cool stuff there soon! https://t.co/2Il8jivyNM
The week's wrapped up, and it looks like it's time to relax! π΄π
If you find it hard to rest because you feel like you're wasting time, try thinking of rest as part of the work - without good rest, your productivity is going to drop!
And part of our team is already kicking back π By the way, one of them is our head of NN - can you guess who? π
@BullTheoryio arrested in turkey, not by the fbi. also the attacks hit canada and us targets too - this wasn't just a european operation.
the Ivanka angle is real but it's one thread in something much bigger
@vijaykaushik93 Honestly, X pays incredibly little, because only Premium (and above) engagement is counted. Compared to something like YouTube, itβs about 100 times less
@AshCrypto green tape's real but the mediator's pakistan, not qatar - sharif and naqvi did the tehran shuttle, qatar's a supporting act. the $400B is just ~0.6% of a $60T market on iran-deal hopes
@AltcoinPsycho defi lending didn't just cool, it halved - TVL down 49% since october. but the capital didn't leave crypto, it rotated straight into your "institutional rails" bucket. RWA protocols were one of the only categories green in Q1.
Same money, better collateral