Hayden Austin

Software engineer & gardener

Hayden Austin.

I build business applications and tools for working with information—from agronomy field records and audio-enhanced books to machine-learning models and GPU execution.

View general CV Explore the work

01 / Selected work

Selected work.

Applications for everyday work, tools for making information usable, and independent research into how models learn and run.

01 / Business software

Equipment platform & custom CMS

Full-stack development · React, Node.js & Firebase

An equipment catalogue for customers, with publishing and administration tools for the people maintaining it.

I built the public React interface, product and article editors, backend data services, staff permissions, reporting and release checks.

Equipment categories with product images and descriptions
Public catalogue · equipment categories
For customers
Browse equipment categories and product information, then find the relevant support and contact routes.
For editors
Maintain products, prepare articles, choose images and publish content through the CMS.
For administrators
Manage staff access, import records, export data and review site activity.
View equipment platform
Engineering details

The implementation separates editor components, data services and permission checks. Rich-text content is sanitised and prepared with slugs and search metadata. Server rendering serves the public pages; automated release gates check tests, access rules and artifact integrity before deployment.

02 / Business software

Nursery operations & content platform

Full-stack development · TypeScript, React & Cloud Functions

A nursery catalogue connected to the everyday work of maintaining plant information, prices, stock and enquiries.

I developed the typed content models, reusable administration editors and service layers, connecting the public catalogue to product, media, publishing and customer-enquiry workflows.

Plant catalogue cards with photographs, care profiles and stock status
Public catalogue · plant profiles and stock information
Keep stock current
Staff can edit product and stock records in a spreadsheet-style interface or import changes in bulk.
Publish with control
Prepare guides and news as drafts, set publication windows and reuse images from a shared library.
Handle exceptions
Per-record results identify bulk edits that need attention; stale-write checks flag conflicting product updates.
View nursery platform
Engineering details

React and TypeScript separate form state from domain services. Validation checks imports and writes; query-cache updates refresh the interface after edits. Backend functions handle enquiries and privileged operations. Media-reference checks support asset management, while previews and unsaved-change guards support editors.

03 / Mobile & environmental data

Agronomy field-record application

In development · React Native, Expo, SQLite & Fastify

An agronomy application for recording rainfall and temperature, keeping corrections, and consulting forecasts alongside field observations.

I developed the mobile record workflows, local persistence and server synchronisation, alongside weather-data validation, caching and visualisation.

Records can be saved without a connection and uploaded later. Corrections retain their history; missing readings stay distinct from zero rainfall.

Data handling & development status

SQLite records double as a durable upload queue. Stable identifiers support retries, dependencies keep corrections after their original records, and conflicts remain visible. TypeScript contracts connect the Expo app to Fastify and PostgreSQL services.

Open-Meteo forecasts are validated and cached with source and freshness information. Failed refreshes preserve previous data. The application is in development; hosted rollout, hardware integration and trained forecasting models are outside the demonstrated scope.

Machine learning

Applied AI & model research.

04 / Applied AI & document engineering

From PDF to a book you can hear

In development · Python, OCR, speech alignment & EPUB

An authoring pipeline for turning PDFs into EPUBs with recorded or generated narration and synchronised text highlighting.

I developed document-conversion components, text-fragment processing, narration alignment and EPUB media-overlay packaging. The reader supports previewing the output; the conversion pipeline is the centre of the project.

Conversion, narration & development status

Fixed-layout and reflowable conversion paths use PDF extraction and optional OCR. Existing recordings follow a Whisper-family transcription and alignment path; generated narration uses Edge TTS. Fragment identifiers and SMIL timing references connect text passages to embedded audio.

The components are implemented, with integration and validation still in progress. Layout preservation, alignment quality and playback compatibility are active development work.

05 / Independent research

How should a model remember?

Research engineering · PyTorch, JAX, CUDA & Triton

I investigate how neural models retain useful information, learn structure and execute efficiently.

I implement experimental model families, learned-memory mechanisms and the training and evaluation code around them. I also build GPU kernels and model-specific execution plans to study the cost of running those ideas.

Questions behind the code

  1. What should memory keep?

    Learned addressing, updates and retrieval.

  2. What structure helps a model learn?

    Geometric layers and matched comparisons.

  3. What does it cost to run?

    Parallel execution, gradients and GPU profiling.

Architecture research · connected questions, distinct experiments

Matched baselines and reference checks help separate model behaviour from implementation effects. I compare learning quality, runtime and memory use, and document where an approach helps and where it does not.

Explore the research code
Architectures, execution & evidence

The monorepo contains separate programmes for associative scans, learned memory, equivariant models and controlled benchmarks. Implementations include CUDA/Triton kernels, custom autograd and shape- and precision-aware dispatch. Output and gradient checks compare optimised paths with references; pinned data and artifact manifests keep results traceable. Fused gathered-memory inference is evaluated separately from trainable model paths.

Research programmes ↗Learned memory & retrieval ↗Compiler implementation ↗Controlled benchmarks ↗

Further application work

Other ways of making information useful.

Analytics application

Strategy-game analytics

I built a React/TypeScript dashboard for comparing game entities and exploring historical snapshots through interactive charts.

Batched API requests, Web Workers for computation, D3 visualisations and coordinated cache invalidation across browser tabs.

Frontend & internationalisation

IntegerSpark

A lightweight calculator with localised pages and number handling across 12 languages, including right-to-left presentation.

I implemented decimal arithmetic, locale-aware input and display, and static page generation with asset-size checks.

Earlier projects & coursework Java, Windows applications & image classification

Java / server tooling

Plugins & custom interfaces

My Java plugin work connects server events to custom item behaviour and inventory interfaces. Resource-pack parsing and persistent metadata let those behaviours use configured assets and item state.

Resource-pack plugin ↗Item plugin ↗

C# / coursework & prototype

Desktop & data access

The WinForms coursework manages student and module records through ADO.NET and SQL Server. A separate WPF prototype wraps command execution with asynchronous output and log display.

WinForms coursework ↗WPF prototype ↗

Applied ML · coursework

Image classification & evaluation

A TensorFlow/Keras CNN with image preprocessing, a stratified training/validation split, confusion-matrix diagnostics and saved-model inference.

Classification notebook ↗

02 / Experience & background

Experience
& education.

Hayden Austin

Software, like a garden, takes balance and patience: structure to support it, room to grow, and regular care as conditions change. I enjoy the work of tending both.

Independent work

Independent software & ML projects

Mobile field-record tools, audio-enhanced document pipelines, analytics applications and independent model research.

Explore independent projects

2023 – 2025

Freelance Developer

Development of customer-facing websites and the administration systems behind business operations.

2019 – 2022

Game Server Developer & Administrator

Event-driven Java plugins and server API integrations, Linux administration, user support and investigation of performance issues.

Education

Belgium Campus ITversity

Studies toward a Bachelor of Computing, specialising in Software Engineering.

03 / Technical depth

Technical skills.

Explore the capabilities behind the projects, and how they connect across applications, research and software systems.

Explore the connected skills 51 nodes · 3 areas

Select a skill for context and follow its connections. Project and source links are included where useful.

Software engineeringThree connected areas of work
Explore application projects ↗
Explore ML & research ↗
Explore supporting work ↗

Tools & applications Shared practices

04 / Curriculum vitae

Download my CV.

One-page PDFs. Choose the general overview or a version tailored to the role.

Get in touch

Let’s build something useful.

For software-engineering roles, application projects or ML research work. Email is the simplest way to reach me.

haydaust@gmail.com GitHub