AIBuildAI-2: A Knowledge-Enhanced Agent for Automatically Building AI Models

AIBuildAI-2 introduces a knowledge-enhanced AI agent that automatically builds machine learning models by combining large language models with an external, evolving knowledge system. The system achieves state-of-the-art performance, ranking first on MLE-Bench and placing in the top 6.6% of human teams in a predictive competition, democratizing AI model development for non-specialists.

AIBuildAI-2 addresses a critical bottleneck in AI development: the gap between demand for machine learning solutions and the scarcity of expertise required to build them. Traditional approaches rely solely on parametric knowledge embedded in language models, which becomes outdated and lacks practical engineering depth. This research represents a meaningful shift toward autonomous model development by introducing a hybrid architecture that augments LLMs with dynamically curated external knowledge.

The hierarchical knowledge system operates in two layers—high-level topical instructions and low-level implementation documents—that the agent selectively retrieves based on task context. Critically, this system evolves through experience, distilling successful runs into structured lessons that feed back into the knowledge base. This creates a feedback loop that improves over time, unlike static LLM parameters.

The competitive validation is significant: achieving top performance on MLE-Bench and ranking within the top 6.6% of 4,370 expert human teams demonstrates that automated approaches now approach human-level performance on complex modeling tasks. This has immediate implications for scientific research, where domain experts in biology, physics, and chemistry increasingly need high-performing models without building in-house ML engineering teams.

The broader impact extends to workforce dynamics and accessibility. By reducing the manual engineering overhead, AIBuildAI-2 lowers barriers to entry for scientific discovery in computational fields. However, the system's effectiveness depends on knowledge curation quality, suggesting emerging opportunities for knowledge engineering specialists rather than replacement of traditional roles.

• →AIBuildAI-2 combines LLMs with external, evolving knowledge systems to automate end-to-end AI model development with state-of-the-art performance.
• →The hierarchical knowledge architecture dynamically loads context-relevant expertise and learns from each completed task to improve future runs.
• →Competitive validation shows the agent outperforms 93.4% of human expert teams on predictive modeling tasks, indicating significant progress toward autonomous development.
• →The system democratizes access to advanced AI model building for domain scientists without specialized ML engineering expertise.
• →Knowledge system architecture creates feedback loops that improve performance over time, addressing the staleness problem of static parametric knowledge in LLMs.