Built an Autonomous Swing Trading Pipeline with Self Validation

Hey everyone, I've been learning AI agents for a few months now and wanted to build something beyond the usual chatbot combining my some of my of skills from swing trading. I built an autonomous agent that runs a swing trading pipeline every day using hermes ai. This was mostly a side project for fun, learning, collecting data and seeing how the strategies actually performed over time. The pipeline screens over 1300+ stocks, logs signals, tracks performance across T+2/+5/+10 and self-validates its own strategy changes every 48 hours. The pipeline screens 1000+ stocks every morning (S&P500, Nasdaq 100, S&P 400), logs the signals, and then tracks every single one at T+2, T+5 and T+10 to see if the call was actually right. Over a couple of months i was able to log 683 signals for tracking. The overall win rate is around 60% at T+10, nothing crazy. But the interesting part was the breakdown. Setup Signals Win Rate Avg Win Avg Loss Mean Reversion (RSI < 35) 94 69.1% +7.7% -6.6% Breakout (RSI < 55) 589 58.9% +6.0% -5.9% Overall 683 60.3% +6.2% -6.0% Mean reversion at 69% win, breakout at 59%. So here's where the self validation loop comes in. Every 48 hours, the system runs a retrospective on all track signals, looks for patterns, and if it suggests a parameter change, it logs that change with a date. Then it waits for 15+ new signals to come in post-change, comapres win rates before and after and if it didn't improve at least 2%, it flags it to revert back to the original strategy. Basically stops me from chasing noise and pretending it's strategy refinement. On a side note, this works great with hermes as it also learns on the fly. So one real example: the system found that RSI 70+ entries were winning 46% of the time compared to 70% for RSI under 40. That drove tightening the breakout RSI from 75 to 55 over the past couple months. Each step got validated before I kept it. This was mostly a side project, paper/shadow tracked not live capital. Repo is here try it out if you want and any feedback is appreciated! Repo in comments.