Qgentic-AI
An automated machine-learning engineering stack in which large-language-model agents take a problem description, produce a plan, generate and run code locally, and iteratively refine a solution, aimed today at Kaggle-style competitions.
Qgentic-AI runs a main agent and a researcher subagent that read a session-wide goal and task-specific instructions from local markdown files, then plan, write code, execute it and analyze results in a loop, writing ideas, research reports and per-iteration outputs under a task directory as they go. The main agent bootstraps the rest of its own working files, and a competition can be launched by its Kaggle identifier, which triggers a dataset download.
Setup requires a CUDA-enabled GPU, Python 3.12, and API keys for the language-model providers, a web-crawling service and Hugging Face used during a run, plus separate Kaggle credentials and a downloaded Kaggle metadata dataset for retrieving competition information. The stack is described as extensible to goals beyond competitions.
It is written in Python.