locked in on AI agents & LLMs, building quant systems onchain.
on chain
Joined February 2026
- Tweets3K
- Following428
- Followers4.9K
- Likes17.5K
Pinned Tweet
The complete playbook for running a quant desk in 2026: tools, stack and the structure that keep you alive.
Someone just published a blueprint for building a market simulator with Grok Bots.
It covers the matching engine, order book, noise traders, value traders, and market makers. All of it.
this is what serious market microstructure engineering actually look
someone open-sourced an AI agent that builds and backtests trading strategies for free
describe the idea → it writes the code → runs the backtest → exports it to TradingView
genuinely one of the best things I’ve seen open-sourced this year
Drop this into Grok and get a real quant research lab.
It builds the pipeline every trading idea has to pass: 12 hypotheses, seven gates, a ledger of every attempt, and a reviewer checking the work.
Now you have it
Paste this into Grok Bot tonight and let it work while you sleep.
By morning it can build you a market simulator: order book, four trader types, experiments run across 30+ seeds each, with a validator checking the work.
It’s all yours
built a full trading agent system on top of NautilusTrader, explained the whole thing from scratch
agents write proposals, a gate filters bad ones out, the engine runs the test and code computes the results.
here’s what that system looks like:
a friend just built a full trading strategy pipeline using AI agents and put the prompt out for free.
five agent roles. six strategy families. truncation tests, stress tests, multiple-testing adjustments.
this is what institutional-level strategy research looks like when agents run it
I’VE NEVER SEEN AN OPEN-SOURCE TRADING FRAMEWORK ENGINEERED THIS SERIOUSLY.
a complete multi-agent LLM trading framework
agents analyse, debate and review risk before every decision
here’s how I’d build a strategy factory around agents like these:
someone just mapped out a full multi-agent trading system spec and made it public.
every agent has one job. every output is typed and structured. every failure is traceable to a specific layer.
this is what institutional AI trading infrastructure looks like in the article below
HEDGE FUNDS BUILD THIS INTERNALLY AND CHARGE MILLIONS TO ACCESS IT
someone built a full ML trading curriculum and put it on GitHub for free
27 chapters that takes you from raw market data all the way to a running strategy
deep learning, causal ML, RL, autonomous agents, all with real working code
21k+ people already have it bookmarked
machine-learning-for-trading
github.com/stefan-jansen/mac…
a single system prompt that turns Claude into a quantitative research engineer
it writes the thesis, drafts the rules, stress-tests them with the tools you connect, and logs every experiment in a versioned record
this is what a disciplined algo research workflow looks like now
a quant researcher can now build a complete strategy playbook using Claude Opus 5.5.
the AI writes the rules, tests on real data, stress tests every parameter and labels every result. DATA VALIDATED. TRADE-OFF. NO EDGE.
every number comes from a tested data layer. this is how serious systematic research actually works.
I’ve seen people spend months trying to understand how quants think.
It comes down to one thing.
They write a rule, test it on real data, and let the numbers decide.
probably that’s the whole edge.
This is the most underrated way to learn how quants think.
They don't guess. They write a simple rule, test it on real market data, and let the numbers decide.
Now you can do the same, step by step, even as a complete beginner.
honestly one of the most quietly powerful repos in all of crypto.
CCXT has been powering crypto trading systems for years. one library, 100+ exchanges, every major language supported.
the unglamorous backbone of the whole algo trading world.
