About the role
Maincode builds frontier AI models end to end. We design the models, run the infrastructure, and train on data we source and shape ourselves. What we have been calling a “data scientist” does not actually exist in the market in the form we need. The work we are doing is frontier work. It changes continuously as models, methods, and training regimes evolve, and there is no stable body of prior experience that reliably predicts success.
This role is best understood internally as a Signal Engineer.
The real work here is not analytics, dashboards, or big data engineering.
It is about how AI models learn, and specifically how the quality, composition, timing, and structure of input signal shapes learning outcomes across the full training lifecycle, from early pre training through mid training, post training, supervised fine tuning, and reinforcement learning.
The analogy we use internally is a fuels engineer for a race car. The engine matters, but performance is ultimately unlocked by obsessing over the fuel. What goes in, how it is blended, when it is introduced, and how it changes under different conditions.
What you would actually do
You would be responsible for designing, sourcing, shaping, and evaluating input signal at scale so that models learn faster, learn better, and generalize more effectively.
This includes:
Working deeply with raw and synthetic data
Designing new datasets and signal mixtures
Iterating on input distributions and formats
Understanding how small changes in signal affect training behavior
Thinking across stages of training, not in isolation
Treating data as a dynamic, evolving system rather than a static asset
This is not about knowing the right answer up front. It is about immersion, experimentation, and developing intuition through long feedback loops.
The kind of person who does well here
Success in this role is driven almost entirely by cognitive traits and working style, not prior AI experience.
People who tend to thrive here:
Have extreme attention to detail and care deeply about precision
Are comfortable with long arcs of repetitive, detailed work
Enjoy staring at complex, noisy information until structure emerges
Are excited by tuning, blending, and calibrating inputs
Have high stamina and patience
Learn new technical domains very quickly when given dense material
Get intrinsic satisfaction from incremental improvement and deep understanding
These are people who do not get bored easily. They metabolize boredom into insight. They are depth driven rather than novelty driven.
We explicitly do not require prior experience with large language models or modern AI training. No one meaningfully has that experience yet. What matters is learning gradient, range, and willingness to throw yourself into the problem space.
How you would work
You would spend your time inside computational environments and technical tooling.
You do not need to identify as a software engineer, but you should be comfortable:
Working directly with data through scripts, notebooks, and experimental pipelines
Modifying existing systems and building small utilities to explore questions
Using code as a tool for thinking, testing, and refinement
The goal is speed of learning and depth of understanding, not polish.
Backgrounds that often fit well
Rather than advertising this role directly, we look for people in fields that reward signal sensitivity, precision, and long horizon reasoning.
We expect to see strong candidates from areas such as:
Actuarial science
Forensic accounting or forensic finance
Quantum computing
Quantitative risk modeling
Applied statistics and signal processing
Econometrics
Operations research
We also see excellent fits from applied mathematics, computational physics, systems engineering, bioinformatics, reliability engineering, and similar disciplines.
Many people in these fields are constrained by conservative or slow moving industries and are actively looking for something more dynamic, even if they do not yet realize there is a place for them in frontier AI.
How we interview
We are explicitly hiring for learning gradient, not recall.
Our process is designed to:
See how quickly you build a mental model of an unfamiliar system
Understand how you reason about signal, feedback, and failure modes
Observe how deeply you engage with input quality as a lever
Learn whether you enjoy the work itself, not just the outcome
Correctness matters less than depth, curiosity, and systems intuition.
The core message
We are offering a path for people who love signal, precision, and deep work to apply that obsession to frontier AI.
This role is not about having the right background. It is about having the right brain, the right stamina, and the desire to become excellent at something that does not quite have a name yet.
Compensation Range: A$150K - A$180K