Emergent Cognition Lab

Building Autonomous AI Research Agents

The Vision

We’re building fully autonomous AI agents capable of performing independent research in Machine Learning and Data Science. Our vision is to create AI systems that can independently propose hypotheses, design experiments, execute them, analyze results, and iterate—pushing the boundaries of scientific discovery without human intervention.

Modern ML research is becoming unmanageable. Researchers must run hundreds of experiments, track countless metrics, correlate code changes with performance improvements, and synthesize insights from an ever-expanding literature. The bottleneck isn’t compute—it’s human attention and analysis capacity.

Research Tools

Query

Coming Soon

Query is an AI-powered experiment management platform that helps ML researchers analyze metrics, code diffs, and results to cluster runs, generate insights, and suggest next steps. Today’s platforms only track data; ours understands it.

query.ai/project/transformer-experiments

Training Metrics

Accuracy
94.2%
↑ 2.3%
Loss
0.142
↓ 0.031
Epoch
48/50
2h 14m left
Learning Rate
0.001
constant

Key Code Changes(commit a3f8b2)

def attention_layer(self, x):
- heads = 8
+ heads = 16 # Increased attention heads
return multi_head_attention(x, heads)
Q
AI Analysis

I've analyzed your last 50 runs. Cluster #3 shows the best performance with 94.2% accuracy. The key difference is the increased attention heads (8→16) in commit a3f8b2.

Recommended Next Steps:
  • →Try doubling batch size to 128 for faster convergence
  • →Test learning rate decay schedule
Foundational Research

Research Publications

Our research explores the foundations of autonomous AI agents for scientific discovery. We investigate agentic workflows, experiment design, hypothesis generation, and how to train AI systems to conduct meaningful research independently.