I build small, verifiable systems — mostly Python, mostly around LLMs.

On every closed candle it renders the price action into a 1920×1080 chart, sends the image with 79 Numba-compiled indicators, the news context and its own closed-trade memory to a multimodal model, then passes the answer through deterministic risk gates before booking a paper trade. The prompt, the retrieved memory, the verdict and the losses are public.

Code · Live dashboard · Ledger · How one cycle works

• ti_numba — 79 technical indicators in Numba JIT, the ones the bot calls every cycle.
• candledownloader — OHLCV downloader with gap detection across exchanges.
• telegram-commandcode — MCP server that lets a coding agent send Telegram messages and files.

Paper trading only. No fund management, no signal selling, and no performance claim that the public ledger cannot reproduce.