Using AutoVoila for Automated Research
Paras Chopra is the founder of Lossfunk, an independent lab exploring foundational questions in biological and artificial intelligence. He previously founded Wingify, and currently serves as the PI for researchers at Lossfunk. His independent research currently focuses on AI and Consciousness, including a position paper accepted at ICML 2026 main conference.
In this hands-on session, Paras will walk through AutoVoila, Lossfunk's open-source autonomous research prompt designed for beginner and student researchers in AI/ML. He will show how to set it up, and how the system takes a starting topic to a finished manuscript draft running autonomously for hours.
We will look at the core design choices behind AutoVoila: the research-philosophy grounding how the agent reasons, the voila markdown file for human-in-the-loop control, and multi-agent review with Codex that grounds the work done by Claude Code. Paras will also share lessons from running these systems in practice and the failure modes researchers should watch for.
The session will end with a discussion of CAISc, including problems attendees can try AutoVoila on across the verifiable and open-ended tracks.
We strongly recommend that all attendees to go through the repo below in detail before the session.
AutoVoila repo — https://github.com/paraschopra/autovoila
Part of the CAISc 2026 pre-conference series. The deadline for submission is May 30, 23:59 AOE. Learn more at caisc2026.github.io.