Software and computing are central to modern experimental physics — from writing detector algorithms and training deep learning models to building analysis pipelines and visualising results. This page collects my technical work across research software, data science, and personal development projects.
01
Research software spanning detector-level C++ algorithms, deep learning inference pipelines, and web development — including work actively deployed in physics analyses at BESIII.
Developing a neutral hadron tagger for the BESIII experiment using a transformer-based deep neural network trained on EMC shower shape and TOF timing signals to distinguish K⁰L, neutrons, and antineutrons. The trained model is exported to ONNX format and integrated into a dedicated BOSS algorithm that reads the ONNX file at runtime and applies the model weights event-by-event — making the tagger directly available to physics analyses at BESIII. Preliminary AUC > 0.95 across all hadron classes.
End-to-end machine learning pipelines for high-energy physics analyses — from raw detector output to trained classifiers. Pipelines cover data ingestion, feature engineering on ROOT ntuples, model training (PyTorch / scikit-learn), hyperparameter optimisation, and performance evaluation with physics-motivated metrics.
Designed and built entirely from scratch using HTML, CSS, and vanilla JavaScript — no frameworks. Features a dark-mode academic aesthetic, bilingual English/Arabic support with full RTL layout, responsive design, and a globe-icon language switcher. Developed through prompt engineering with Claude Code (Anthropic). Hosted on GitHub Pages.
02
Formal training in data science through the Udacity Nanodegree, applied to real-world datasets and documented in open-source projects on GitHub.
Advanced programme (~4 months, 10 hrs/week) assuming prior fluency in Python, statistics, and machine learning. Five courses: Solving Data Science Problems — CRISP-DM process, EDA, communicating results (project: data science blog post); Software Engineering for Data Scientists — clean & modular code, unit testing, OOP, Flask/Plotly dashboards; Data Engineering — ETL pipelines, NLP (tokenisation, tf-idf, sentiment analysis), scikit-learn ML pipelines with grid search (project: disaster response message classifier on real crowdsourced data); Experiment Design & Recommendations — A/B testing, multiple comparison corrections (Bonferroni, FDR, Tukey), matrix factorisation & FunkSVD (project: recommendation engine on IBM Watson Studio); Capstone — choice of dog breed classification with CNNs, Starbucks customer behaviour modelling, Arvato financial services segmentation, or Spark big-data churn prediction on AWS/IBM Cloud.
Portfolio
Projects available on GitHub — real-world dataset analysis, predictive modelling, data visualisation, and NLP tasks. All include documented notebooks with reproducible results.
03
Exploring cross-platform mobile development as a way to bring scientific tools and data-driven interfaces into everyday reach.
A cross-platform Flutter book reader supporting both PDF and EPUB formats. Library Management — import, store, and delete files with persistent storage and per-book reading progress tracking. PDF Reader — pinch-to-zoom viewing via the pdfx package. EPUB Reader — richer reading experience with Text-to-Speech chapter narration (Flutter TTS), on-device English → Spanish translation via Google ML Kit (works fully offline), table-of-contents navigation, and interactive text selection for instant translation. Planned addition: AI-assisted paper summarisation and technical reading support.