Date: Thursday, November 6, 2025
Location: Bloomberg Student Center, Room 204 (Hybrid)
Registration: Register here
Colab Notebook: Open in Colab
Submission Form: Google Form
AtomGPT.org API (AGAPI): GitHub Repository
YouTube video https://www.youtube.com/watch?v=L2iqlogVLAc
The Agentic AI for Science (AAI4Science) Hackathon 2025 invites participants to explore and evaluate the capabilities of AI tools across scientific disciplines.
This event aims to benchmark how well agentic AI systems reason, explain, and generalize across Physics, Chemistry, Mathematics, and Biology (PCMB) tasks.
These models can generate insightful and accurate responses, but they also hallucinateโproducing incorrect outputs with high confidence.
Understanding, evaluating, and mitigating such behaviors is essential for trustworthy and reproducible scientific discovery.
- Gain hands-on experience with cutting-edge AI tools.
- Identify and categorize hallucinations (false but confident outputs).
- Develop and share reproducible test prompts and datasets.
- Foster open collaboration toward benchmarking AI for science.
| Time | Activity |
|---|---|
| 1:00 pm โ 1:10 pm | Welcome & Orientation |
| 1:10 pm โ 1:15 pm | Deanโs Remarks |
| 1:15 pm โ 3:00 pm | Hackathon Session |
| 2:50 pm โ 3:00 pm | Closing Discussion |
Format: Hybrid (in-person + Zoom)
A Zoom link will be emailed to all registered participants.
Refreshments will be provided for in-person attendees.
- Join the event in person or virtually.
- Attend the introduction session.
- Open the Colab notebook to explore examples.
- Complete the assigned tasks using the AI tools provided.
- Save your Colab notebook as a GitHub Gist.
- Add your Colab link (with solutions) in the Google Form.
- Submit your results through the form before the deadline.
-
Top evaluations and insights will be highlighted on the JARVIS Leaderboard
-
Get featured in a future journal publication summarizing the hackathon outcomes.
-
Win gift cards
Dr. Kamal Choudhary
Assistant Professor, Johns Hopkins University
Email: kchoudh2@jhu.edu
Website: AtomGPT.org
Letโs make AI for science reliable, transparent, and truly agentic!