Hackathon 4 - Building agents for STEM - Slautin

Hackathon 4 - Building agents for STEM - Slautin

by Machine Learning in the Nanoworld

Formal & Physical Sciences AI — News & Applications UYComputer scienceUYQArtificial intelligence (AI)
🎙 Machine Learning in the Nanoworld 👥 1K 📅 July 18, 2026 ⏱ 12 min 👁 9 📄 tutorial 🧭 2026-08-16
Available in: English (current) Français

Keywords

smolagentsGeminiSTEMimage analysistools

Summary

This hackathon session, led by Slautin, guides participants through building and adjusting a simplistic AI agent using Hugging Face’s smolagents framework with Gemini as its reasoning core. The primary objective is to extend the agent’s Python-based toolset to better analyze STEM images, specifically by developing tools that allow the agent to accurately detect and classify atomic columns. The session begins with an introduction to the notebook and the smolagents framework, followed by instructions on setting up the environment, including installing libraries and uploading STEM images. Participants are shown how to create and register tools using decorators, and how to configure the agent with Gemini, including obtaining an API key. The agent is then tested with simple tasks like loading an image and reporting signal-to-noise ratio, and more complex tasks like estimating the number of atoms, which reveals the need for additional tools. The main task for participants is to extend the toolset to accurately count and classify atomic columns in images, particularly for a barium copper oxide sample. The session emphasizes hands-on experimentation and understanding the limitations and advantages of such agent-based approaches.

183 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides practical value by demonstrating a concrete implementation of an AI agent for STEM image analysis. It offers a clear workflow for setting up the environment, creating tools, and using the agent. The argumentation is based on the demonstration of the agent’s capabilities and limitations, supported by examples. However, the video does not delve into theoretical foundations or compare with alternative approaches, limiting its depth. The open-ended task encourages exploration, which is valuable for learning.

Scientific Rigor, Source Quality, Title Accuracy

The video does not cite any external sources, and the description provides no links. The scientific rigor is moderate: the approach is reproducible, but the video lacks detailed explanations of the algorithms or validation of results. The title accurately reflects the content. No comments were provided for analysis.

140 words

Title / Content Match

The title accurately reflects the content: a hackathon session focused on building agents for STEM applications.

Quality & Reliability

6/10

The video is a practical tutorial demonstrating the use of Hugging Face's smolagents framework with Gemini for STEM image analysis. It provides clear step-by-step instructions and code examples, but lacks in-depth explanation of underlying concepts and does not cite external sources. The content is reproducible and the approach is sound, but the scientific depth is limited.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Contribution & Novelties

The video provides a hands-on tutorial for building AI agents for STEM image analysis, specifically using smolagents and Gemini. It demonstrates the process of extending agent capabilities through custom tools, which is a practical skill. The open-ended task encourages innovation.

Pour aller plus loin :

77 words

Radar Profile

The radar profile shows moderate scores across all dimensions, indicating a balanced but not exceptional tutorial. The highest score is in information quantity and quality, reflecting the practical content, while technical level and reliability are slightly lower due to lack of depth and external validation.

Reliability 6/10