On April 24th, we hosted an AI Deep Dive Session in collaboration with the Bastos Lab and the 91吃瓜网 Brain Institute with from 91吃瓜网’s Department of Psychology in the College of Arts and Science, where we explored a multi-agent LLM framework designed to semi-automate two of the most time-intensive components of neuroscience research: scientific literature review and data analysis. Dr. Bastos, whose lab investigates the neural mechanisms of prediction, attention, and working memory using large-scale neuronal recordings and computational modeling, led a discussion on pushing agentic systems beyond summarization into rigorous, reasoning-driven scientific analysis.
Watch the AI Deep Dive:
Highlights:
- Purpose: Architect multi-agent LLM systems capable of interpreting high-level scientific prompts, selecting analytical methods, executing them against real datasets, and feeding results back into a self-improving research loop.
- Focus Areas: Closing the loop between literature synthesis and quantitative analysis, building reusable analysis skills, and designing evaluation strategies for scientific trustworthiness.
- AI Applications: Multi-agent LLM pipelines, modular reasoning skills (such as burst detection for identifying learning signatures in neural data), and agentic patterns that produce reproducible neuroscience workflows from natural language prompts.
Session Insights:
- The conversation pressure-tested how to extend agentic systems from summarization into genuine reasoning over scientific data, with particular attention to methodological trust, validation, and the risk of plausible-but-wrong outputs.
- Collaboration with the Bastos Lab grounded the discussion in concrete neuroscience workflows, surfacing what reusable analysis “skills” need to look like to translate high-level prompts into reliable, reproducible pipelines.
- This work serves as foundational material for Dr. Bastos’s Genesis grant proposal and the upcoming MaDeLaNe workshop hosted at DSI in June, with clear opportunities for further collaboration on architecture, evaluation, and scientific accuracy.
Conclusion:
The AI Deep Dive with Dr. Bastos and the Bastos Lab showcased how agentic AI systems can move beyond surface-level summarization to support rigorous, reproducible scientific discovery in neuroscience. This session provided a unique opportunity for those interested in agentic AI, neuroscience, and scientific reasoning to engage in meaningful discussion and collaboration.
Are you interested in hosting a future AI Deep Dive? Contact us at datascience@vanderbilt.edu.