Primus is an autonomous AI researcher that automates the entire research loop—hypothesizing, literature review, coding, experiment execution, and paper writing—running up to 30× faster than human teams. It can handle tasks such as fine‑tuning models, writing optimized kernels, extending existing papers, and exploring new domains, delivering complete research artifacts and published‑ready results.
Funding
Funding not disclosed
Founders
Product
Problem
Conducting scientific research requires a sequential loop of hypothesis generation, literature review, coding, experiment execution, and manuscript writing, each step typically performed by human experts and often taking weeks or months to complete.
Solution
Primus is an autonomous AI system that executes the full research loop without human intervention, operating up to 30 times faster than traditional teams. It generates hypotheses, scans millions of papers, writes code, runs experiments on real compute resources, iterates based on results, and produces complete research manuscripts. Users can request tasks such as model fine‑tuning, kernel optimization, paper extension, framework porting, or domain exploration, and Primus delivers experimental outcomes together with a finished paper. The platform continuously improves as it conducts more experiments, enabling rapid discovery across diverse fields including machine learning, physics, biology, and more.
Target Audience
Primus targets machine‑learning researchers, data scientists, and engineering teams who need rapid prototyping, optimization, or exploratory studies without allocating large human research resources.
Features
- End‑to‑end automation of hypothesis generation, literature synthesis, coding, experiment execution, and manuscript drafting
- Ability to process and cite millions of scientific papers to inform research directions
- Real‑world compute execution with support for GPU‑accelerated workloads and custom kernel generation
- Iterative learning loop that refines approaches based on experimental results to improve future performance
- Multi‑domain capability allowing tasks such as model fine‑tuning, kernel optimization, framework porting, and novel domain exploration
- Output of complete, publishable research papers including experimental data, analysis, and code artifacts