aqbor888/AlphaGPT
🤖 Enhance your trading strategies with AlphaGPT, a powerful tool for navigating decentralized markets and optimizing liquidity with ease.
What's novel
🤖 Enhance your trading strategies with AlphaGPT, a powerful tool for navigating decentralized markets and optimizing liquidity with ease.
Code Analysis
20 files read · 4 roundsA Solana meme coin trading bot that uses a small Transformer trained with REINFORCE to generate factor formulas for entry/exit signals, fetches trending tokens from Birdeye, executes trades via Jupiter DEX, and provides a Streamlit dashboard for monitoring.
Strengths
The project has a coherent end-to-end pipeline from data ingestion through factor mining to live execution, with reasonable risk management (stop-loss, trailing stops, honeypot checks, liquidity filters). The LoRD regularization experiment in lord/experiment.py shows genuine research thinking about attention weight structure.
Weaknesses
The README is completely dishonest about what the code does (describes a Chinese stock consumer app when it's a Solana meme coin bot). There are zero tests, a hardcoded API token in times.py, the RL training loop is naive (no action masking in model_core unlike times.py, no proper advantage estimation), and the 'AlphaGPT' model is a 64-dim 2-layer Transformer that's barely more than a toy.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
73
Contributors
4
Files
56
Active weeks
4
Repository
Language
Python
Stars
1
Forks
4
License
Apache-2.0