PhAI Labs Released ScienceBuddy Research Workspace
The platform enables researchers to refine AI models through integrated toolsets and feedback loops.
Updated on Sept. 24, 2026 in Artificial Intelligence
Live Poll
Do you trust AI research tools that actively learn from user feedback?
PhAI Labs has launched ScienceBuddy Preview, a browser-based workspace integrating scientific papers and analysis tools to improve model performance through iterative research feedback. The research-stage system is now accessible to the public.
Why it matters
The system seeks to accelerate AI development by formalizing the feedback loop between researcher interaction and model training. It addresses the growing need for collaborative environments where model outputs are continuously validated against scientific tasks.
The workspace hosts 224 tools across 22 functional modules. Evaluation of the system showed a 31.1% gain in model accuracy, with 33.3% of tasks moving from incorrect to correct results.
The players
PhAI Labs
A research-focused AI company that develops specialized tools for scientific computing and model iteration.
The details
The platform employs a two-loop learning process that adjusts both the task model and the software harness—a set of automated tools that execute and verify code. By capturing research session data, the system converts interactions into new evaluation criteria, which then inform subsequent training cycles. The workspace supports both English and Chinese languages for its integrated suite of tools.
Timeline
September 24, 2026: PhAI Labs released the ScienceBuddy Preview.
The Tech Race
The platform joins a field of research-focused environments aiming to reduce the friction in AI model iteration. It follows the pattern set by open science protocols to bridge the gap between human inquiry and automated evaluation.
The workspace is available as a browser-based tool at science-buddy.io for researchers and developers. Users can utilize the interface in English or Chinese to integrate papers and data into their current scientific workflows.
The takeaway
The system represents a shift toward treating researcher interaction as a primary dataset for model improvement. Users can watch for future updates on GitHub, where PhAI Labs has published the research code under the MIT License.
Further reading
For broader trends in research tooling, explore our coverage of Artificial Intelligence.
More information
To access the platform or review its capabilities, visit the ScienceBuddy workspace access.
Live Poll
Do you trust AI research tools that actively learn from user feedback?
