Since agent processing take significant time, what happens to this latency induced if agentic-ai is implemented in existing workflow. What are the latency challenges ? What could be key strategies to overcome challenges? What should we do to change the user expectation.=? What should be done to maintain/enhance user experience? What trade-offs should be considers between performance, latency, cost etc?

Recent breakthroughs in large language model-based artificial intelligence (AI) have captured the public’s interest in AI more broadly. With the growing adoption of these technologies in professional and educational settings, public dialog about their potential impacts on the workforce has been ubiquitous. It is, however, difficult to separate the public dialog about the potential impact of the technology from the experienced impact of the technology in the research software engineer and data science workplace. Likewise, it is challenging to separate the generalized anxiety about AI from its specific impacts on individuals working in specialized work settings.

As research software engineers (RSEs) and those in adjacent computational fields engage with AI in the workplace, the realities of the impacts of this technology are becoming clearer. However, much of the dialog has been limited to high-level discussion around general intra-institutional impacts, and lacks the nuance required to provide helpful guidance to RSE practitioners in research settings, specifically. Surprisingly, many RSEs are not involved in career discussions on what the rise of AI means for their professions.

During this BoF, we will hold a structured, interactive discussion session with the goal of identifying critical areas of engagement with AI in the workplace including: current use of AI, AI assistance and automation, AI skills and workforce development, AI and open science, and AI futures. This BoF will represent the first of a series of discussions held jointly by the Academic Data Science Alliance and the US Research Software Engineer Association over the coming year, with support from Schmidt Sciences. The insights gathered from these sessions will inform the development of guidance resources on these topic areas for the broader RSE and computational data practitioner communities.

The rapid growth of scientific data repositories demands innovative solutions for efficient metadata creation. In this talk, we present our open-source project that leverages large language models to automate the generation of standard-compliant metadata files from raw scientific datasets. Our approach harnesses the capabilities of pre-trained open source models, finetuned with domain-specific data, and integrated with Langgraph to orchestrate a modular, end-to-end pipeline capable of ingesting heterogeneous raw data files and outputting metadata conforming to specific standards.

The methodology involves a multi-stage process where raw data is first parsed and analyzed by the LLM to extract relevant scientific and contextual information. This information is then structured into metadata templates that adhere strictly to recognized standards, thereby reducing human error and accelerating the data release cycle. We demonstrate the effectiveness of our approach using the USGS ScienceBase repository, where we have successfully generated metadata for a variety of scientific datasets, including images, time series, and text data.

Beyond its immediate application to the USGS ScienceBase repository, our open-source framework is designed to be extensible, allowing adaptation to other data release processes across various scientific domains. We will discuss the technical challenges encountered, such as managing diverse data formats and ensuring metadata quality, and outline strategies for community-driven enhancements. This work not only streamlines the metadata creation workflow but also sets the stage for broader adoption of generative AI in scientific data management.

Additional Material: - Project supported by USGS and ORNL - Codebase will be available on GitHub after paper publication - Fine-tuned LLM models will be available on Hugginface after paper publication

Flyte is a Linux Foundation OSS orchestrator built for Data and Machine Learning workflows focused on scalability, reliability, and developer productivity. Flyte’s Python SDK, Flytekit, empowers developers by shipping their code from their local environments onto a cluster with one simple CLI command. In this talk, you will learn about the design and implementation details that powers Flytekit’s core features, such as “fast registration” and “type transformers”, and a plugin system that enables Dask, Ray, or distributed GPU workflows.

The increasing prevalence of AI models necessitates robust mechanisms to ensure their trustworthiness. This talk introduces a standardized, PKI-agnostic approach to verifying the origins and integrity of machine learning models, as built by the OpenSSF Model Signing project. We extend this methodology beyond models to encompass datasets and other associated files, offering a holistic solution for maintaining data provenance and integrity.

Real-time machine learning depends on features and data that by definition can’t be pre-computed. Detecting fraud or acute diseases like sepsis requires processing events that emerged seconds ago. How do we build an infrastructure platform that executes complex data pipelines (< 10ms) end-to-end and on-demand? All while meeting data teams where they are–in Python–the language of ML! Learn how we built a symbolic interpreter that accelerates ML pipelines by transpiling Python into DAGs of static expressions. These expressions are optimized in C++ and eventually run in production workloads at scale with Velox–an OSS (~4k stars) unified query engine (C++) from Meta.

AI, particularly generative AI, is rapidly transforming the scientific landscape, offering unprecedented opportunities and novel challenges across all stages of research. This Birds of a Feather session aims to bring together researchers, developers, and practitioners to share experiences, discuss best practices, and explore the evolving role of AI in science.

The Universe isn't always so quiet: neutron stars, fast radio bursts, and potentially alien civilizations emit bursts of electromagnetic energy - radio transients - into the unknown. In some cases, these emissions, like with pulsars, are constant and periodic; but in others, like with fast radio bursts, they're short in duration and infrequent. Classical detection surveys typically rely on dedispersion techniques and human-crafted signal processing filters to remove noise and highlight a signal of interest. But what if we're missing something?

In this talk we will introduce a workflow to avoid classical processing all together. By feeding RF samples directly from the telescope's digitizers into GPU computing, we can train an AI model to serve as a detector -- not only enabling real time performance, but also making decisions directly on raw spectrogram data, eliminating the need for classical processing. We will demonstrate how each step of the pipeline works - from AI model training and data curation to real-time inferencing at scale. Our hope is that this new sensor processing architecture can simplify development, democratize science, and process increasingly large amounts of data in real time.

Camera traps are an essential tool for wildlife research. Zamba is an open source Python package that leverages machine learning and computer vision to automate time-intensive processing tasks for wildlife camera trap data. This talk will dive into Zamba's capabilities and key factors that influenced its design and development. Topics will include the importance of code-free custom model training, Zamba’s origins in an open machine learning competition, and the technical challenges of processing video data. Attendees will walk away with a better understanding of how machine learning and Python tools can support conservation efforts.

In today’s world of ever-growing data and AI, learning about GPUs has become an essential part of software carpentry, professional development and the education curriculum. However, teaching with GPUs can be challenging, from resource accessibility to managing dependencies and varying knowledge levels.

During this talk we will address these issues by offering practical strategies to promote active learning with GPUs and share our experiences from running numerous Python conference tutorials that leveraged GPUs. Attendees will learn different options to how to provide GPU access, tailor content for different expertise levels, and simplify package management when possible.

If you are an educator, researcher, and/or developer who is interested in teaching or learning about GPU computing with Python, this talk will give you the confidence to teach topics that require GPU acceleration and quickly get your audience up and running.

Women remain critically underrepresented in data science and Python communities, comprising only 15–22% of professionals globally and less than 3% of contributors to Python open-source projects. This disparity not only limits diversity but also represents a missed opportunity for innovation and community growth. This talk explores actionable strategies to address these gaps, drawing from my leadership in Women in AI at IBM, TechWomen mentorship, and initiatives with NumFOCUS. Attendees will gain insights and practical steps to create inclusive environments, foster diverse collaboration, and ensure the scientific Python community thrives by unlocking its full potential.

Generative AI has rapidly changed the landscape of computing and data education. Many learners are utilizing generative AI to assist in learning, so what should educators do to address the opportunities, risks, and potential for their use? The goal of this open discussion session is to bring together community members to unravel these pressing questions in order to not only improve learning outcomes in a variety of diverse contexts: not only students learning in a classroom setting, but also ed-tech or generative AI designers developing new user experiences that aim to improve human capacities, and even scientists interested in learning best practices for communicating results to stakeholders or creating learning materials for colleagues. The open discussion will include ample opportunity for community members to network with each other and build connections after the conference.

Collaborating on code and software is essential to open science—but it’s not always easy. Join this BoF for an interactive discussion on the real-world challenges of open source collaboration. We’ll explore common hurdles like Python packaging, contributing to existing codebases, and emerging issues around LLM-assisted development and AI-generated software contributions.

We’ll kick off with a brief overview of pyOpenSci—an inclusive community of Pythonistas, from novices to experts—working to make it easier to create, find, share, and contribute to reusable code. We’ll then facilitate small-group discussions and use an interactive Mentimeter survey to help you share your experiences and ideas.

Your feedback will directly shape pyOpenSci’s priorities for the coming year, as we build new programs and resources to support your work in the Python scientific ecosystem. Whether you’re just starting out or a seasoned developer, you’ll leave with clear ways to get involved and make an impact on the broader Python ecosystem in service of advancing scientific discovery.

NVIDIA’s CUDA platform has long been the backbone of high-performance GPU computing, but its power has historically been gated behind C and C++ expertise. With the recent introduction of native Python support, CUDA is more accessible to the programming language you know and love, ushering in a new era for scientific computing, data science, and AI development.

Synthetic aviation fuels (SAFs) offer a pathway to improving efficiency, but high cost and volume requirements hinder property testing and increase risk of developing low-performing fuels. To promote productive SAF research, we used Fourier Transform Infrared (FTIR) spectra to train accurate, interpretable fuel property models. In this presentation, we will discuss how we leveraged standard Python libraries – NumPy, pandas, and scikit-learn – and Non-negative Matrix Factorization to decompose FTIR spectra and develop predictive models. Specifically, we will review the pipeline developed for preprocessing FTIR data, the ensemble models used for property prediction, and how the features correlate with physicochemical properties.

AI assistants are evolving from simple Q&A bots to intelligent, multimodal, multilingual, and agentic systems capable of reasoning, retrieving, and autonomously acting. In this talk, we’ll showcase how to build a voice-enabled, multilingual, multimodal RAG (Retrieval-Augmented Generation) assistant using Gradio, OpenAI’s Whisper, LangChain, LangGraph, and FAISS. Our assistant will not only process voice and text inputs in multiple languages but also intelligently retrieve information from structured and unstructured data. We’ll demonstrate this with a flight search use case—leveraging a flight database for retrieval and, when necessary, autonomously searching external sources using LangGraph. You will gain practical insights into building scalable, adaptive AI assistants that move beyond static chatbots to autonomous agents that interact dynamically with users and the web.

Designing tomorrow's materials requires understanding how atoms behave – a challenge that's both fascinating and incredibly complex. While machine learning offers exciting speedups in materials simulation, it often falls short, missing vital electronic structure information needed to connect theory with experimental results. This work introduces a powerful solution: Density Functional Tight Binding (DFTB), which, combined with the versatile tools of Scientific Python, allows us to understand the electronic behavior of materials while maintaining computational efficiency. In this talk, I will present our findings demonstrating how DFTB, coupled with readily available Python packages, allows for direct comparison between theoretical predictions and experimental data, such as XPS measurements. I will also showcase our publicly available repository, containing DFTB parameters for a wide range of materials, making this powerful approach accessible to the broader research community.

This talk explores various methods to accelerate traditional machine learning pipelines using scikit-learn, UMAP, and HDBSCAN on GPUs. We will contrast the experimental Array API Standard support layer in scikit-learn with the cuML library from the NVIDIA RAPIDS Data Science stack, including its zero-code change acceleration capability. ML and data science practitioners will learn how to seamlessly accelerate machine learning workflows, highlight performance benefits, and receive practical guidance for different problem types and sizes. Insights into minimizing cost and runtime by effectively mixing hardware for various tasks, as well as the current implementation status and future plans for these acceleration methods, will be provided.

This talk presents a candid reflection on integrating generative AI into an Engineering Computations course, revealing unexpected challenges despite best intentions. Students quickly developed patterns of using AI as a shortcut rather than a learning companion, leading to decreased attendance and an "illusion of competence." I'll discuss the disconnect between instructor expectations and student behavior, analyze how traditional assessment structures reinforced counterproductive AI usage, and share strategies for guiding students toward using AI as a co-pilot rather than a substitute for critical thinking while maintaining academic integrity.

Explainable AI (XAI) emerged to clarify the decision-making of complex deep learning models, but standard XAI methods are often uninformative on Earth system models due to their high-dimensional and physically constrained nature. We introduce “physical XAI,” which adapts XAI techniques to maintain physical realism and handle autocorrelated data effectively. Our approach includes physically consistent perturbations, analysis of uncertainty, and the use of variance-based global sensitivity tools. Furthermore, we expand the definition of “physical XAI” to include meaningful interactive data analysis. We demonstrate these methods on two Earth system models: a data-driven global weather model and a winter precipitation type model to show how we can gain more physically meaningful insights.

At CERN (European Organization for Nuclear Research), machine learning models are developed and deployed for various applications, including data analysis, event reconstruction, and classification. These models must not only be highly sophisticated but also optimized for efficient inference. A critical application is in Triggers- systems designed to identify and select interesting events from an immense stream of experimental data. Experiments like ATLAS and CMS generate data at rates of approximately 100 TB/s, requiring Triggers to rapidly filter out irrelevant events. This talk will explore the challenges of deploying machine learning in such high-throughput environments and discuss solutions to enhance their performance and reliability.

X-ray ptychographic imaging is becoming an indispensable tool for visualizing matter at nanoscale, driving innovation across many fields, including functional materials, electronics, life sciences, etc. This imaging mode is particularly attractive thanks to its ability to generate high-resolution view of an extended object without using a lens with high numerical aperture. The technique relies on advanced mathematical algorithms to retrieve the missing phase information that is not directly recorded by a physical detector, therefore computation intensive. Advances in accelerator, optics, and detector technologies have greatly increased data generate rate, imposing a big challenge on efficient execution of reconstruction process to support decision-making in an experiment. Here, we demonstrate how efficient GPU-based reconstruction algorithms, deployed at the edge, enable real-time feedback during high-speed continuous data acquisition increasing the speed and efficiency of the experiments. The developments further pave the way for AI-augmented autonomous microscopic experimentation performed at machine speeds.

Generative Artificial Intelligence (AI) is reshaping engineering education by offering students new ways to engage with complex concepts and content. Ethical concerns including bias, intellectual property, and plagiarism make Generative AI a controversial educational tool. Overreliance on AI may also lead to academic integrity issues, necessitating clear student codes of conduct that define acceptable use. As educators we should carefully design learning objectives to align with transferrable career skills in our fields. By practicing backward design with a focus on career-readiness skills, we can incorporate useful prompt engineering, rapid prototyping, and critical reasoning skills that incorporate generative AI. Engineering students want to develop essential career skills such as critical thinking, communication, and technology. This talk will focus on case studies for using generative AI and rapid prototyping for scientific computing in engineering courses for physics, programming, and technical writing. These courses include assignments and reading examples using NumPy, SciPy, Pandas, etc. in Jupyter notebooks. Embracing generative AI tools has helped students compare, evaluate, and discuss work that was inaccessible before generative AI. This talk explores strategies for using AI in engineering education while accomplishing learning objectives and giving students opportunities to practice career readiness skills.

Scaling artificial intelligence (AI) and machine learning (ML) workflows on high-performance computing (HPC) systems presents unique challenges, particularly as models become more complex and data-intensive. This study explores strategies to optimize AI/ML workflows for enhanced performance and resource utilization on HPC platforms.​

We investigate advanced parallelization techniques, such as Data Parallelism (DP), Distributed Data Parallel (DDP), and Fully Sharded Data Parallel (FSDP). Implementing memory-efficient strategies, including mixed precision training and activation checkpointing, significantly reduces memory consumption without compromising model accuracy. Additionally, we examine various communication backends( i.e. NCCL, MPI, and Gloo) to enhance inter-GPU and inter-node communication efficiency. Special attention is given to the complexities of implementing these backends in HPC environments, providing solutions for proper configuration and execution.​

Our findings demonstrate that these optimizations enable stable and scalable AI/ML model training and inference, achieving substantial improvements in training times and resource efficiency. This presentation will detail the technical challenges encountered and the solutions developed, offering insights into effectively scaling AI/ML workflows on HPC systems.​

Block-based programming divides inputs into local arrays that are processed concurrently by groups of threads. Users write sequential array-centric code, and the framework handles parallelization, synchronization, and data movement behind the scenes. This approach aligns well with SciPy's array-centric ethos and has roots in older HPC libraries, such as NWChem’s TCE, BLIS, and ATLAS.

In recent years, many block-based Python programming models for GPUs have emerged, like Triton, JAX/Pallas, and Warp, aiming to make parallelism more accessible for scientists and increase portability.

In this talk, we'll present cuTile and Tile IR, a new Pythonic tile-based programming model and compiler recently announced by NVIDIA. We'll explore cuTile examples from a variety of domains, including a new LLAMA3-based reference app and a port of miniWeather. You'll learn the best practices for writing and debugging block-based Python GPU code, gain insight into how such code performs, and learn how it differs from traditional SIMT programming.

By the end of the session, you'll understand how block-based GPU programming enables more intuitive, portable, and efficient development of high-performance, data-parallel Python applications for HPC, data science, and machine learning.