@_zheergen: Insane! HKU lab has directly open-sourced a mini quant hedge fund. Vibe-Trading, 34K+ Stars, MIT open-source, a financi…

Insane! HKU lab has directly open-sourced a mini quant hedge fund.

Vibe-Trading, 34K+ Stars, MIT open-source, a financial research platform that can source data on its own, mine factors, run backtests, and assemble Agent teams.

In the past, building a quant strategy meant sourcing data yourself, writing factors, picking a backtest engine, tuning parameters, reviewing results—the whole process pieced together bit by bit by hand.

Vibe-Trading is different: you just say in natural language what you want to research, and it handles the rest—data, factors, backtests, Agent division of labor—all running on its own.

Core capabilities:

1️⃣ 90 Finance Skills: Agents can directly call on tools for market data, options, factors, backtesting, research, and more

2️⃣ 462 ready-made Alphas: Qlib158, Alpha101, GTJA191, academic factors, fundamental factors—all neatly organized

3️⃣ 10+ backtest engines: A-shares, global stocks, Crypto, futures, forex, options—multiple markets ready to test

4️⃣ 30 Swarm Teams: Investment committees, Quant Desks, Risk Reviews, Macro, Crypto—teams you can spin up for collaboration on the spot

5️⃣ Persistent Memory: Past research, sessions, skills, and context carry over—no starting from scratch every time

6️⃣ Shadow Account: Upload your broker journal, and AI reverse-engineers your trading patterns for counterfactual backtests and audits

What I find most intriguing about it is that AI is finally moving beyond just “answering financial questions.”

It’s starting to truly execute an entire research chain:

You pose a question, it sources data on its own, runs analysis, does backtests, convenes Agents for a meeting, and finally hands back the research results.

Work that used to take a whole table of quant researchers is now kicking off with a single sentence.

If you’re into AI quant, stock research, Crypto, factor mining, or want to actually plug Agents into your investment research workflow, this one’s absolutely worth bookmarking.

爱吃折耳根的Ace (@_zheergen): 我去！做股票量化机器学习，最浪费时间的往往不是训练模型，而是每天手搓一堆因子。

tsfresh，9.4K Stars，MIT 开源，一套专门把“时间序列特征工程”自动化的 Python 工具。

以前做因子研究，要自己想波动率、峰值、趋势、偏度、频域、相关性，再一个个写代码验证。

tsfresh 不一样，它先从 K