Senior Research Scientist

The Onset; Sydney; On-site; AUD 180,000 - 240,000; Full time

Job summary

The Onset is an early‑stage Australian AI research company building foundational intelligent systems. You will lead original research in memory, reasoning, continual learning and model training, with freedom to shape a research agenda and ship working prototypes.

Reporting to the Chief Scientist, you will publish work at top venues and mentor PhD interns while helping turn ideas into demonstrators and platforms. Australia‑based applicants preferred.

Qualifications

• PhD in ML/AI/CV/NLP is required.
• Strong publication record at respected AI/ML venues.
• Ability to independently take research from a question to a defensible result.
• Deep expertise in memory, reasoning, continual learning, planning, multimodal learning or large‑scale model training.
• Strong implementation skills and willingness to stay close to the code.

Responsibilities

• Define and lead original research across memory, reasoning, planning, continual learning, multimodal AI, simulation or large‑scale model training.
• Explore new model and system architectures for intelligent systems that operate and adapt long term.
• Take research from question to experimentation, validation and publication.
• Publish at leading venues such as NeurIPS, ICML, ICLR, CVPR and ECCV.
• Collaborate with machine learning and systems engineers to turn research into demonstrators and platform capabilities.
• Develop rigorous approaches to simulation, evaluation and measurement.
• Mentor PhD interns and junior researchers.
• Remain hands‑on with implementation and AI‑assisted development tools.

Skills

Publication record

Independent research

Memory & reasoning

Continual learning

Multimodal systems

Large-scale model training

Academic collaboration

Education

PhD in ML/AI/CV/NLP

Job description

“I’m publishing research, but I rarely get to see what happens when it meets the real world.”

“I want to work on something more fundamental than another layer around an existing model.”

“I’m ready to own a research direction, not inherit a narrow piece of someone else’s roadmap.”

“I still want to publish, but I also want to build.”

If any of this resonates, please keep reading.

An early-stage Australian AI research company is working on a difficult question.

How do you build intelligent systems that can keep working, learning and making good decisions long after the initial prompt?

Most current AI systems have no durable understanding of what has happened before. Each new interaction relies on rebuilding context from prompts, transcripts or stored files. That can work for short, contained tasks. It becomes far less reliable when a system needs to operate for days, months or years, balance competing priorities and learn from the consequences of its decisions.

This team is developing the models and underlying infrastructure needed to change that.

The research spans memory, reasoning, continual learning and model adaptation. Alongside it, the engineering team is building a runtime that can coordinate specialised AI processes, manage resources, respond to urgent events and preserve the state of work over time.

The goal is to create intelligent systems that can operate continuously in complex environments, including robotics and industrial operations, where forgetting context or mishandling priorities has genuine consequences.

This is not another chatbot or a thin agent framework built around an existing model.

The work sits closer to the foundations: new research, new architectures and the systems required to make them useful.

You’ll join a small, founder-led team of researchers and engineers who publish papers and ship code. Reporting to the Chief Scientist, you’ll have the freedom to shape an original research agenda while helping turn promising ideas into working systems.

Day to day

• Define and lead original research across memory, reasoning, planning, continual learning, multimodal AI, simulation or large-scale model training.
• Explore new model and system architectures for intelligent systems that need to operate and adapt over long periods.
• Take research from the initial question through experimentation, validation and publication.
• Publish at leading venues such as NeurIPS, ICML, ICLR, CVPR and ECCV.
• Work closely with machine learning and systems engineers to turn research into demonstrators and platform capabilities.
• Develop rigorous approaches to simulation, evaluation and measurement.
• Collaborate with external researchers and the broader academic community.
• Mentor PhD interns and more junior researchers.
• Remain hands‑on with implementation, experimentation and AI‑assisted development tools.

“I’m publishing research, but I rarely get to see what happens when it meets the real world.”

“I want to work on something more fundamental than another layer around an existing model.”

“I’m ready to own a research direction, not inherit a narrow piece of someone else’s roadmap.”

“I still want to publish, but I also want to build.”

If any of this resonates, please keep reading.

An early-stage Australian AI research company is working on a difficult question.

How do you build intelligent systems that can keep working, learning and making good decisions long after the initial prompt?

Most current AI systems have no durable understanding of what has happened before. Each new interaction relies on rebuilding context from prompts, transcripts or stored files. That can work for short, contained tasks. It becomes far less reliable when a system needs to operate for days, months or years, balance competing priorities and learn from the consequences of its decisions.

This team is developing the models and underlying infrastructure needed to change that.

The research spans memory, reasoning, continual learning and model adaptation. Alongside it, the engineering team is building a runtime that can coordinate specialised AI processes, manage resources, respond to urgent events and preserve the state of work over time.

The goal is to create intelligent systems that can operate continuously in complex environments, including robotics and industrial operations, where forgetting context or mishandling priorities has genuine consequences.

This is not another chatbot or a thin agent framework built around an existing model.

The work sits closer to the foundations: new research, new architectures and the systems required to make them useful.

You’ll join a small, founder-led team of researchers and engineers who publish papers and ship code. Reporting to the Chief Scientist, you’ll have the freedom to shape an original research agenda while helping turn promising ideas into working systems.

Day to day

• Define and lead original research across memory, reasoning, planning, continual learning, multimodal AI, simulation or large-scale model training.
• Explore new model and system architectures for intelligent systems that need to operate and adapt over long periods.
• Take research from the initial question through experimentation, validation and publication.
• Publish at leading venues such as NeurIPS, ICML, ICLR, CVPR and ECCV.
• Work closely with machine learning and systems engineers to turn research into demonstrators and platform capabilities.
• Develop rigorous approaches to simulation, evaluation and measurement.
• Collaborate with external researchers and the broader academic community.
• Mentor PhD interns and more junior researchers.
• Remain hands‑on with implementation, experimentation and AI‑assisted development tools.

Success here will not be measured by the number of people you manage.

It will be measured by the quality of the research you originate, your ability to test and implement it, and whether you can move it towards both publication and a functioning product.

Ideal background

• A PhD in machine learning, artificial intelligence, computer vision, NLP or a closely related field.
• A strong publication record at respected AI and machine learning venues.
• Evidence that you can independently take research from a question to a defensible result.
• Deep expertise in at least one of memory, reasoning, continual learning, planning, multimodal learning, simulation or large‑scale model training.
• Strong implementation skills and a willingness to remain close to the code.
• Experience in a high-quality research lab, deep‑tech company or ambitious AI startup.
• Comfort working with ambiguity, testing ideas quickly and changing direction when the evidence demands it.
• An ability to collaborate closely while still challenging assumptions and forming your own view.

This probably won’t suit someone looking to step away from technical work, manage a large research function or inherit a fully specified roadmap.

It may suit someone who wants genuine intellectual ownership, strong academic peers and the chance to help shape a new technical architecture while the important decisions are still being made.

The team is distributed and meets in person regularly. Australia is preferred, although exceptional international candidates will be considered.

Here, you’ll get to do good research, build models and see whether the ideas actually work.

For a confidential discussion, contact Ronny Haughton at *****@theonset.com.au or 0448 808 ***.