Project 02 / 10Scientific visualization · Research software

IonTrace

University of Sydney · Jan 2026 – Jun 2026

A browser-based, AI-assisted GUI for loading, reconstructing and exploring atom probe tomography datasets.

Completed
Software
Research

IonTrace's Exploration step: ion species toggles, point size and render mode controls on the left, and a 3D point cloud of 3,000 atoms in the viewport — Fig. 01IonTrace's Exploration step, shown with a sample dataset.

Key result

APT file formats loaded

Delivered

Supported formats are .rhit, .vval, .apt, .pos, .epos, .rng and .rrng. A display-budget system keeps the 3D view responsive on datasets of hundreds of thousands of ions.

Technical highlights

A five-step guided workflow in the browser: upload, reconstruction, ranging, exploration and analysis.
AI-suggested reconstruction parameters, each with a confidence label, plus automatic mass-range intervals.
A display-budget system that dynamically down-samples the point cloud so the 3D view stays responsive on very large datasets.

My contribution: Built IonTrace as a research intern at the University of Sydney (Next.js and TypeScript front end, Python backend)
Implemented the display-budget down-sampling for the 3D point cloud
Worked through the full APT workflow: SEM-based specimen preparation on a Xe plasma FIB-SEM, and fabricating specimens from 3D-printed, hardened metal powders
Tools: Next.jsTypeScriptPythonXe plasma FIB-SEM

Overview

IonTrace is a browser-based tool I built as a research intern at the University of Sydney for working with atom probe tomography (APT) data, from raw files to an explorable 3D reconstruction, with AI-assisted parameter suggestions along the way.

Engineering challenge

APT datasets arrive in several file formats and can hold hundreds of thousands of ions. Researchers need one tool that can load them, run the reconstruction and let them explore the result without the 3D view slowing down.

Design process

Alongside the software, I worked through the full APT workflow: SEM-based sample preparation on a Xe plasma FIB-SEM, and fabricating specimens from metal powders by 3D printing and hardening.

IonTrace guides the user through five steps: upload, reconstruction, ranging, exploration and analysis. A Next.js/TypeScript front end sits on a Python backend that parses the multiple APT file formats and runs the reconstruction pipeline.

AI assistance is built into several steps. Reconstruction suggests values for its parameters from similar datasets, with an overall confidence score and High or Medium labels on each value. Ranging can generate mass-range intervals automatically with Auto-Range, and the Analysis metrics (cluster count, average spacing, interface width) carry the same confidence labels. An in-app research assistant sits alongside the workflow.

To keep the viewport responsive on very large datasets, I implemented a display-budget system that dynamically down-samples the point cloud. The user chooses how many points to render, and the interface shows the atoms currently visible, a level-of-detail setting and an estimated frame rate.

In the Exploration step you can toggle ion species, adjust point size and switch between points, density, clusters and iso-surface render modes.

Testing & results

A working tool that loads, reconstructs and explores APT datasets, and stays responsive on datasets with hundreds of thousands of ions. The screenshots on this page use a sample dataset.

Gallery

IonTrace's Upload step with a file drop area listing the supported formats and a points-to-render setting — Fig. 02Upload: drop in APT files (.rhit, .vval, .apt, .pos, .epos, .rng, .rrng) and choose how many points to render.

IonTrace's Reconstruction step with an 89 percent AI confidence gauge, the reconstruction model, and parameters such as k-factor, image compression factor, field factor and detection efficiency with High or Medium confidence labels — Fig. 03Reconstruction: pick a model (Bas-Chen by default) and tune parameters, with AI-suggested values and confidence labels.

IonTrace's Ranging step with a mass spectrum, colored ranges for Al, Fe, Ni and Cu, and a bulk composition table — Fig. 04Ranging: a mass spectrum with colored ion ranges, plus the bulk composition.

Fig. 05Analysis: a 1D concentration profile, region-of-interest statistics and composition, with export and report options.

Scanning electron image from the Xe plasma FIB-SEM of a small specimen being prepared, at 2,500 times magnification — Fig. 06Specimen preparation on the Xe plasma FIB-SEM (30 kV, 82.9 µm field of view).

Close-up scanning electron image of a specimen with a 5.936 micrometer measurement marked in green, at 5,000 times magnification — Fig. 07Close-up during preparation: 5,000×, 41.4 µm field of view, with a 5.9 µm feature measured.

Scanning electron image of a small specimen on its support at 2,000 times magnification — Fig. 08The specimen on its support at 2,000× (104 µm field of view).

Four-panel FIB-SEM control view showing electron and ion beam images, the chamber camera, and the EasyLift micromanipulator needle near a specimen — Fig. 09The FIB-SEM control view during specimen preparation, with the EasyLift micromanipulator inserted.