DeepScientist - A fully automated AI scientist system launched by Westlake University

What is DeepScientist?

DeepScientist is an autonomous research system proposed by Westlake University and other institutions, automating the entire process from hypothesis generation to experimental verification, results analysis, and paper writing. Through a multi-agent architecture and reinforcement learning strategies, artificial intelligence can continuously explore, verify, and advance scientific frontiers without human intervention. The system's core employs a closed-loop process of "Hypothesize–Verify–Analyze," combined with Bayesian optimization principles, treating scientific discovery as a problem of finding the optimal solution within a vast methodological space. DeepScientist includes a built-in "Findings Memory," which records and reuses historical experimental results, improving research efficiency and innovation.

DeepScientist's main functions

• Automated research hypothesis generationThe system can autonomously propose new research hypotheses or improvement plans based on existing knowledge and experimental results, thus automating the process from thinking to innovation.
• Experimental Design and ImplementationIt can transform hypotheses into executable experimental procedures or code, automatically build experimental environments, run models, collect data, and verify results.
• Results Analysis and Report GenerationAfter the experiment is completed, the system automatically analyzes the results, summarizes the patterns, and generates academic reports or paper drafts to support the direct output of scientific research results.
• Scientific discoveries in memory management (Findings Memory)Record all experiments, hypotheses, and results to form a reusable knowledge base to assist in decision-making regarding future research directions.
• Autonomous optimization and learningBy using methods such as Bayesian optimization, we can balance exploration and utilization in the research space and continuously improve research efficiency and the quality of results.
• Multi-agent collaborationBy having multiple dedicated AI agents (hypothesis generation, code execution, result analysis, etc.) work together, a safe, controllable, and modular research process can be built.
• Security and verification mechanismsThe system employs a sandbox and containerized execution strategy to ensure experimental safety, reliable results, and automatic verification of experimental conclusions.

DeepScientist's technical principles

• Scientific discovery modeling as an optimization problemThis approach views scientific innovation as a problem of finding the optimal solution within a vast methodological space, and utilizes Bayesian optimization and surrogate models to efficiently evaluate and screen research hypotheses.
• Closed-loop scientific research process mechanismThrough a three-stage cycle of "hypothesis-verification-analysis", scientific hypotheses are continuously generated, tested and improved, forming a self-driven scientific research iteration system.
• Multi-agent architectureThe system consists of multiple dedicated agents, each responsible for strategy planning, code implementation, result analysis, and report generation, working together to complete the entire scientific research process.
• Findings Memory SystemEstablish a long-term knowledge memory to store past experiments, hypotheses, and results to guide new scientific research explorations and optimizations.
• Stratified Validation and Agency EvaluationA two-level verification strategy of low-fidelity and high-fidelity is adopted. Potential is first evaluated through a surrogate model, and then computationally intensive real experiments are conducted to save resources.
• Container and Sandbox Execution EnvironmentRun code and experiments in an isolated and secure environment to prevent system conflicts and misoperations, and to ensure that experiments are reproducible and results are reliable.
• Automatic result verification mechanismAfter the experiment, the system will independently perform verification again to prevent false successes and ensure the reliability and verifiability of scientific conclusions.

DeepScientist project address

• Project official websitehttps://ai-researcher.net
• Github repositoryhttps://github.com/ResearAI/DeepScientist
• arXiv technical paper: https://arxiv.org/pdf/2509.26603

Applications of DeepScientist

• Artificial Intelligence Algorithm ResearchIt can autonomously explore model structure, optimize strategies and training methods, and promote the development of AI inference efficiency, interpretability and robustness.
• Automation scientific research and innovationIt can automatically generate and verify new hypotheses in fields such as machine learning, computer vision, and natural language processing, thereby accelerating the iteration of scientific research results.
• Experimental Science AidIt can be used in disciplines such as physics, chemistry, and biology that require extensive experimental verification, automatically screening potential discoveries through virtual experiments and data analysis.
• Intelligent agent system optimizationIn multi-agent collaborative or reinforcement learning tasks, it automatically improves policies and communication mechanisms to optimize system performance.
• Research process automationIt helps research teams automate the entire process from conception to report generation, improving research efficiency and output quality.
• Academic paper generation and review simulationThe system automatically generates paper drafts based on experimental results and uses an AI review module for self-checking and quality assessment.