AI-Researcher

AI-Researcher automates research tasks for researchers who want to develop an idea into code, experiments and an academic manuscript. It can start with a detailed proposal or use reference papers to suggest a research direction.

• Literature review and ideas: Collects and assesses papers, code and datasets from sources such as arXiv, Google Scholar, GitHub and Hugging Face to inform research proposals.
• Algorithm implementation: Develops methods and working code from your stated requirements or ideas derived from reference papers.
• Experiments and analysis: Tests implementations, evaluates performance, refines algorithms and interprets experimental results.
• Paper drafting: Generates a full-length academic manuscript after the research agent completes its work.

The examples focus on machine learning research, including vector quantization, recommendation systems, diffusion and flow matching, and graph neural networks. A Gradio browser interface provides access to the research workflow.

The Python application runs on your own hardware or server, with a Linux Docker environment for the research agent. It can use GPUs or run without one. Model access uses LiteLLM, with an OpenRouter configuration that calls Google's Gemini models. Those cloud calls require internet access and an API key, and send model requests to the external service; self-hosting the application doesn't make that workflow offline. Novix is a separate hosted offering linked by the project.