Embedded Files
An Automatic Math Rersearch AI Workflow (15/03/2026)
An Automatic Math Rersearch AI Workflow (15/03/2026)
There is a question for me. When can AI start doing math research? Now I have the answer. The latest ChatGPT has ability to generate a PhD level math research, without any technical prompt. See my Github blog.
A Math Reading Agent with Visualization (12/03/2026)
A Math Reading Agent with Visualization (12/03/2026)
GenAI can help mathematicians read articles. However, LLM can not generate good visualization between the text output. If readers ask AI to summarize the articles and generate images separately, the learning process will be disturbed. Therefore, I vibe coded the agentic AI workflow to do these simultaneously, with a refinement feature so that user can prompt a specialized summary.
The Github page. The streamlit app.
Update: Anthropic updated Claude with such feature at night. Exactly same day. I shouldn't waste an afternoon on this.
Vibe Coding Duolingo (04/03/2026)
Vibe Coding Duolingo (04/03/2026)
About neural network in equity and ETF trading (15/02/2026)
About neural network in equity and ETF trading (15/02/2026)
This project applies neural network to trade stocks and ETFs.
For stocks, I select tens of stocks and use transformer to study the correlations of their stock features. The annual return for the trading strategy is 28.4%, while the return for QQQ in the same period is 14.6%.
For ETFs, I only use the data of selected ETF and VIX. I compare buy&hold strategy, the static parameter strategy, and MLP-training parameter strategy. The average 252-trading-day (annual) return of the MLP algorithm is 13.41% (11.97% for buy & hold, 16.98% for static parameter) for SPY, 29.12% (13.35% for buy & hold, 22.27% for static parameter) for QQQ, and 88.20% (12.30% for buy & hold, 36.30% for static parameter) for TQQQ.
The graphs below are the return curves for SPY(left), QQQ(middle), TQQQ(right).
This shows that although NN cannot predict the trend of the price, traders can use it to adjust the ruled-base trading strategies in high-volatity but trending-up ETFs.
Below is a report generated by Claude.
Page updated
Google Sites
Report abuse