About Edexia
Edexia is on a mission to fix the mess we call the education system. We're starting with assessment—the area that places the greatest burden on educators and offers the most immediate impact for positive change. By revolutionising how assessment works, we're creating a cascade of improvements that will transform education from the ground up.
While others in EdTech are content with wrapping a basic UI around existing LLMs, we're taking the opposite approach—building sophisticated AI systems that actually solve the hardest problems in education (look at “What You'll Build” to find out more).
Our vision extends far beyond assessment. We're creating the foundation for systemic educational change, addressing the fundamental misalignments and inefficiencies that have plagued schools for generations. We've already secured contracts with schools across Australia and the US. We raised $4M USD at a $35M valuation after completing Y Combinator.
About Us
We're a small, focused 3-person team based in Brisbane, Australia. The CEO, Daniel Gibbon, and CTO Nathan are both second-time technical founders who previously built an EdTech to $200k ARR at 18 in just 1 year while achieving a perfect GPA. Now we are all on a mission to rebuild the education system from the ground up with AI.
This Job is Perfect for You If
You're excited about solving challenging technical problems that have direct, meaningful impact on education and are willing to work relentlessly toward a central vision.
You value being part of a highly intelligent, driven team that pushes each other to excel.
Must Haves
Be incredibly smart, good at solving complex problems, and learning new things quickly.
Be willing to work relentlessly to accomplish a vision of rebuilding our education system from the ground up with AI.
Genuinely care about having a positive impact through your career.
Have good baseline life and work habits and a willingness to relentlessly optimise every aspect of your life to maximise your productivity and fulfilment.
Have some form of technical background, even if that is just math or physics.
Have the confidence to keep pace with some of the smartest, most driven people in the world.
Be incredibly smart, good at solving complex problems, and learning new things quickly.
Be willing to work relentlessly to accomplish a vision of rebuilding our education system from the ground up with AI.
Genuinely care about having a positive impact through your career.
Have good baseline life and work habits and a willingness to relentlessly optimise every aspect of your life to maximise your productivity and fulfilment.
Have some form of technical background, even if that is just math or physics.
Have the confidence to keep pace with some of the smartest, most driven people in the world.
Nice to Haves
Domain expertise in software engineering.
Domain expertise in AI/ML.
Domain expertise in education.
Historical achievement that backs up your intellectual abilities (e.g. IMO/IOI medals).
Domain expertise in software engineering.
Domain expertise in AI/ML.
Domain expertise in education.
Historical achievement that backs up your intellectual abilities (e.g. IMO/IOI medals).
What You'll Build
As part of our team, you'll work on aspects of these challenges that both:
Move the needle most significantly for our company and mission
Align with your interests and strengths—whether that's core AI architecture, UI/UX design, reinforcement learning systems, or data pipeline engineering
Move the needle most significantly for our company and mission
Align with your interests and strengths—whether that's core AI architecture, UI/UX design, reinforcement learning systems, or data pipeline engineering
Assessment is our strategic starting point. To provide significant impact in this area, we need to solve two core technical challenges, after which we can continue to work on the infinite number of technical challenges to ultimately rebuild education. The two core current technical challenges are:
- The Human-AI Alignment Problem: Interpreting Rubrics
Problem: Grading rubrics include vague terms like "informed" versus "adequate," resulting in inconsistent, inaccurate, and biased evaluation from both humans and LLMs. We need to extract teachers' existing conscious and unconscious interpretation of these variables and help them further define them into precise evaluation parameters.
Solutions:
Rubric unpacking workflow to make a sophisticated initial guess of the interpretation
Real-time voice AI rubric training coaches that have conversations with teachers to align on interpretation
Reinforcement learning components that extract patterns from teacher corrections
Rubric unpacking workflow to make a sophisticated initial guess of the interpretation
Real-time voice AI rubric training coaches that have conversations with teachers to align on interpretation
Reinforcement learning components that extract patterns from teacher corrections
- The Complex Marking Process: Accurately Evaluating Student Work
Problem: Assessment requires making hundreds of discrete decisions about student work that could be in any format (typed, handwritten, graphs, equations) and cover any subject area. Each parameter of analysis could be evaluating anything from mathematical validity to grammar to logical coherence.
Multi-agent workflows that break down complex assessment tasks into processing, identification, analysis, and communication
OCR and computer vision systems that can handle diverse formats
Task-specific analysis modules for different types of evaluation
Systems for converting AI judgments into clear, actionable feedback
Multi-agent workflows that break down complex assessment tasks into processing, identification, analysis, and communication
OCR and computer vision systems that can handle diverse formats
Task-specific analysis modules for different types of evaluation
Systems for converting AI judgments into clear, actionable feedback
Both challenges require a sophisticated two-level optimization approach:
System architecture level: Breaking down complex problems into flexible multi-agent infrastructures that subdivide the work intelligently
End-to-end evaluation frameworks to test complete performance
Individual agent level: Optimising each specific component for its particular task
Unit-level testing for each component and decision point
When optimising individual agents, we often do the following:
First, we evaluate and combine existing models on the market in intelligent ways
We experiment with different task decompositions and prompting techniques
When needed, we explore fine-tuning and reinforcement learning from feedback
As a last resort, we develop our own custom models for specific tasks
First, we evaluate and combine existing models on the market in intelligent ways
We experiment with different task decompositions and prompting techniques
When needed, we explore fine-tuning and reinforcement learning from feedback
As a last resort, we develop our own custom models for specific tasks
To power ongoing improvement, you'll help build our data engine. Like Tesla's autonomous driving fleet, we're creating systems that collect and leverage real-world usage.
Tech Stack
Python for AI development and model integration
Various LLM APIs (OpenAI, Anthropic, etc.)
Custom OCR and computer vision components
Reinforcement learning frameworks
Web technologies for the frontend (handled primarily by our CTO)
Cloud infrastructure for deployment and scaling
Python for AI development and model integration
Various LLM APIs (OpenAI, Anthropic, etc.)
Custom OCR and computer vision components
Reinforcement learning frameworks
Web technologies for the frontend (handled primarily by our CTO)
Cloud infrastructure for deployment and scaling
Zoom Call 1 (~45 minutes): Initial Introductions
Get to know your background and career journey
Please provide an overview of our business model and technical challenges
Share our company vision and current projects
Discuss our AI-powered grading and feedback system
Brief explanation of our interview process and basic logistical considerations
Get to know your background and career journey
Please provide an overview of our business model and technical challenges
Share our company vision and current projects
Discuss our AI-powered grading and feedback system
Brief explanation of our interview process and basic logistical considerations
Zoom Call 2 (~60 minutes): "Is this the best position for you?"
We send you a template spreadsheet to complete before the call
You identify key career factors you care about with relative weightings
You list your potential career pathways/options
During the call, we discussed how well Edexia matches your priorities
We provide detailed context about the role, compensation, and growth opportunities
→ If a tentative yes, we continue to the next step
We send you a template spreadsheet to complete before the call
You identify key career factors you care about with relative weightings
You list your potential career pathways/options
During the call, we discussed how well Edexia matches your priorities
We provide detailed context about the role, compensation, and growth opportunities
→ If a tentative yes, we continue to the next step
Zoom Call 3 (~60 minutes): "Are you the best person for the position?"
We share our rubric outlining what we're looking for in candidates
You can provide feedback on our rubric if you disagree with anything
We ask you to perform self-reflection on how well you match our requirements
Our discussion focuses on technical problem-solving abilities, coding experience, and learning capacity
We assess your fit for solving our advanced AI challenges and system architecture problems
→ If a tentative yes, we continue to the next step
We share our rubric outlining what we're looking for in candidates
You can provide feedback on our rubric if you disagree with anything
We ask you to perform self-reflection on how well you match our requirements
Our discussion focuses on technical problem-solving abilities, coding experience, and learning capacity
We assess your fit for solving our advanced AI challenges and system architecture problems
→ If a tentative yes, we continue to the next step
Zoom Call 4 (~60 minutes): Technical Interview
A deeper technical assessment of skills relevant to our AI systems
Focus on problem-solving approaches rather than just coding
Assessment of your ability to break down complex problems
→ If successful, we will continue to next step
A deeper technical assessment of skills relevant to our AI systems
Focus on problem-solving approaches rather than just coding
Assessment of your ability to break down complex problems
→ If successful, we will continue to next step
1 Week Paid Trial
You work on real projects with our team
Both sides get hands-on experience working together
We evaluate cultural fit and technical capabilities in practice
→ If both questions from Call 2-3 still look like a "yes" then we proceed to final step
You work on real projects with our team
Both sides get hands-on experience working together
We evaluate cultural fit and technical capabilities in practice
→ If both questions from Call 2-3 still look like a "yes" then we proceed to final step
Full-Time Offer