Poster Session

Participants

Poster Title

Abstract

Freilla Mae Espinola

Collaboration Across Python Communities in Asia

This poster presents the a year of initiatives of the Python Asia Organization (PAO), a regional non-profit focused on enabling collaboration between Python communities across Asia. The purpose of PAO is to establish shared structures that help local organizers work together more efficiently while preserving their autonomy and identity.

Key efforts over the past year include: coordinating PythonAsia 2026 in Manila with Python Philippines; launching twice-yearly Online Charity Talks that provide a regional platform for speakers; and forming volunteer workgroups that support essential community operations such as promotions, call for proposal review, infrastructure, and Code of Conduct alignment.

The significance of this work lies in reducing duplicated effort, improving access to resources and cross-border opportunities, and supporting more sustainable community development. By facilitating collaboration rather than centralization, PAO aims to strengthen the resilience and long-term growth of Python communities across Asia.

Sanjiban Sengupta

Fast ML Inference in High-energy Physics using SOFIE

Experiments at CERN, the European Organization for Nuclear Research, produce data at a tremendously high rate of hundreds of GB every second. These data are analysed in real time using statistical and machine learning methods to identify interesting events and discard the rest before the data can be stored for future analysis.

SOFIE (System for Optimized Fast Inference code Emit) is a tool developed at CERN that translates trained ML models into optimized C++ code for inference with BLAS as the sole dependency. The generated code is designed to run in environments requiring low latency, high throughput, and constrained memory. Benchmarking on CPU architectures shows ML inference that is on average 2× faster and uses 50% less memory than ONNX Runtime, considered the state of the art.

SOFIE has also been extended to heterogeneous architectures by integrating the alpaka library, enabling the generated code to run on CPUs and vendor-agnostic GPUs with minimal user changes. This allows inference on co-processors while avoiding host–device–host memory transfers. With SOFIE’s Python interface, both generation and inference can also be executed in a Python environment. In this poster, we present SOFIE, its recent improvements, and its applications in constrained environments such as high-energy physics experiments, autonomous vehicles, and network firewalls.

Rodelene Joy Leonorio

Why Python Works Well for Kids (Visual to Text Coding Transition)

Python offers a simple, readable syntax that helps young learners shift smoothly from block coding to text-based programming, supporting deeper problem-solving habits and clearer logical thinking.

Dima Maharika Dinama

pycon.id : Your python-powered conference platform

We create an open source conference platform using python, which used for PyCon ID

Arcy Layne L. Sace

Development of an Observation Scheduling Tool for Exoplanet Transit Campaigns with the SkyNet Robotic Telescope Network

To support an ongoing transiting exoplanet observation campaign using the SkyNet Robotic Telescope Network [1], an observation scheduling tool is developed as part of a planned end-to-end observation, reduction, and analysis pipeline. The campaign aims to update transit timing measurements of identified exoplanets to maintain accurate ephemerides for planning of observations using both ground- and space-based telescopes.

Given a desired observation period and list of target exoplanets, the tool automatically (1) retrieves data of transit times from the NASA Exoplanet Archive (NEA), (2) identifies the observatories in SkyNet where the transits will be observable within specified observation constraints e.g. airmass and twilight period, (3) generates a comprehensive list of observation times for queuing, (4) creates an iCalendar file for scheduling, and (5) produces airmass plots and sky charts for reference. The tool is developed in Python using astroquery [2] for querying external databases, and astroplan [3] and astropy [4] for data visualization.

Existing online tools from SkyNet and NEA provide airmass plots and support for scheduling, but there are limitations in the number of observatories, targets, and observation window length that can be simultaneously processed. With our observation scheduling tool, users can generate a comprehensive observation plan and maximize the use of the SkyNet Robotic Telescope Network, consisting of 32 telescopes in 20 observatories located in 5 countries across 4 continents. The generated outputs are checked to be consistent with the above-mentioned established online tools.

In addition to using the tool for our own collaboration’s exoplanet campaign, the scheduling tool is planned for open-source release to enable use and further development of the broader astronomy community. Future improvements could include expansion of the observatory database beyond SkyNet, and development of a graphical user interface and web application for increased accessibility for students and citizen scientists.

Mona Obedoza

Analyzing Gender Biases in Gemini AI with Python

With more and more youth turning to generative AI tools for advice and support in their day-to-day lives, I set out to research whether personal advice generated by Google Gemini perpetuates gender biases and stereotypes. I used Python's Gemini API to generate large samples of responses to questions like "I'm a 15-year-old girl and I've been feeling down lately. Any idea what I can do?”, and the NLTK library to visualize and quantify differences in word frequencies between responses to "boy" prompts and responses to "girl" prompts. With the statsmodels package, I found statistically significant differences between hobby and career recommendations given to personas of different genders, and differences in how the AI addresses the notion of feeling/emotion. On a broader scale, I found that Python can accelerate research into the increasingly relevant subject of LLM output and its underlying biases.

Francesca Santiago and Berj Alpeche

Real‑Time Spectrogram Rendering and Audio Buffer Visualization in PySide6

This work presents a PySide6-based program for real‑time visualization of audio waveforms and spectrograms, designed for low-latency rendering and responsive GUI performance. The interface is constructed in Qt Designer and loaded dynamically using QUiLoader, while visualization is implemented with PyQtGraph’s PlotWidget for waveform displays and ImageItem for high‑throughput spectrogram rendering. Audio acquisition and playback are managed via sounddevice streams, with a dedicated OutputStream for WAV sample playback and an InputStream for live microphone capture. Both input and output signals are stored in rolling NumPy buffers using np.roll, enabling constant-time insertion and stable time-domain visualization without blocking audio callbacks. Spectrogram generation is based on a short‑time Fourier transform (STFT) computed over sliding windows with a Hann taper, configurable FFT size, and window overlap. Magnitude spectra are converted to decibels and aggregated into a fixed-length history to produce a continuous, scrolling time-frequency representation. The architecture highlights practical considerations for developing real‑time DSP applications with PySide6, including thread‑safe buffer handling, separation of audio and visualization pipelines, efficient image updates using lookup‑table‑based colormaps, and consistent spectrogram scaling across playback and live microphone modes.