As an AI researcher, over the past few months I’ve been working with Claude Code to help me with my day-to-day research workflows, however, i found its current abilities quite limited when it comes to leveraging existing open-source frameworks (like vLLM, TRL, etc.) to actually run real research workflows.
After Anthropic released the concept of skills, i think this is for sure the right direction for building more capable AI research agents.
If we feed these modularized AI research skills to an agent, i basically empower the agent to actually conduct real AI experiments, including preparing datasets, executing training pipelines, deploying models, and validating scientific hypotheses.
It’s currently a growing library of 43 AI research & engineering skills, covering:
- model pre-training and post-training (RL) workflows
As an AI researcher, over the past few months I’ve been working with Claude Code to help me with my day-to-day research workflows, however, i found its current abilities quite limited when it comes to leveraging existing open-source frameworks (like vLLM, TRL, etc.) to actually run real research workflows.
After Anthropic released the concept of skills, i think this is for sure the right direction for building more capable AI research agents. If we feed these modularized AI research skills to an agent, i basically empower the agent to actually conduct real AI experiments, including preparing datasets, executing training pipelines, deploying models, and validating scientific hypotheses.
It’s currently a growing library of 43 AI research & engineering skills, covering:
- model pre-training and post-training (RL) workflows
- evaluation and visualization
- optimization and inference
data prep, model, dataset, etc.
i'd love your feedbacks :)