Edexia Taught an AI to Grade Like You - Red Pen and All
Two Brisbane engineers built a teaching assistant that learns a teacher's marking style, then hands the judgment back. It is now live in more than 40 schools and running inside Y Combinator's Winter 2025 batch.
Ask any secondary school teacher what they do on Sunday and a surprising number will describe the same scene: a mug going cold, a stack of essays, and a red pen. Marking is the quiet tax on teaching - ten to twenty hours a cycle, invisible to almost everyone outside the staffroom. Edexia was built by two people who watched that tax up close and decided it did not have to be paid in full.
Edexia is an AI teaching assistant that grades essays and written work. The twist, and the reason it is worth paying attention to, is not that it grades - plenty of tools claim that. It is how it grades. Edexia studies a teacher's own marks, learns the standard that particular teacher applies, and then assesses new work the same way. When it gets something wrong, the teacher corrects it and the model adjusts. The point is not to replace the person holding the pen. It is to make a very good copy of their judgment, then hand the pen back.
01 / THE PROBLEMThe weekend that never ends
The founders, Daniel Gibbon and Nathan Wang, did not arrive at this from a whiteboard brainstorm. Both had already built education businesses as teenagers - tutoring operations that reached around $200,000 in annual revenue while they were still in school, both maintaining perfect grades. They knew the market from the inside. What they kept running into was a bottleneck that no amount of good teaching could fix: feedback takes time, and time is the one thing teachers do not have.
The consequence is not just tired teachers. It is slower learning. The most effective loop for a student writer is simple - write, get feedback, rewrite - but at a class of thirty it collapses under the weight of the marking. Edexia's pitch is to make that loop run in a single evening.
Students can write, get instant feedback, and rewrite in an evening. That improvement cycle wasn't possible before. English teacher, St Bernard's College
02 / THE PROOFNumbers a skeptical profession will read
Teachers are, sensibly, hard to convince. So Edexia put a number on it. In a trial across 579 essays at St Bernard's College, its grades matched the teacher's mark exactly 81.2% of the time, and landed within a single mark 98.3% of the time. For a profession where two experienced markers do not always agree, that is a credible claim rather than a flashy one.
03 / HOW IT WORKSShow your working
The most teacherly thing about Edexia is that it shows its working. Rather than returning a grade from a black box, it visually breaks down how it read the rubric - and updates that reading on the fly as the teacher corrects it. That transparency is what turns suspicion into trust. A teacher can see why the machine landed where it did, argue with it, and watch it learn.
Around that core sit the features that make it usable in a real department. There is blind grading, which distributes de-identified scripts at random to strip out bias. There is handwriting transcription for the exam-hall reality that not everything is typed. There are calibration and moderation tools so a whole faculty can mark to one standard, and a library of more than 100 practice prompts filterable by text, theme and difficulty. And crucially, teachers keep the last word: every AI-drafted comment can be edited, rewritten, held for approval, or replaced with a recorded voice note before a student ever sees it. Review mode means nothing goes out that the teacher has not signed off on.
That design choice - AI drafts, human decides - is quietly the whole philosophy. It is easy to build a tool that spits out a grade. It is harder to build one a teacher will actually trust with their students, their standards, and their name on the report. Edexia's expertise is less about the model and more about the classroom: what a rubric really means, how a department moderates, why a mark of 14 instead of 15 can matter to a Year 12 student. The founders' years running education businesses show up in these details.
04 / THE FEARThe bot question, answered
Every English department in 2026 is asking the same nervous question: did a student actually write this, or did a chatbot? Edexia builds the answer into the product. It can replay a piece of writing keystroke by keystroke and attach an AI-likelihood score, giving teachers evidence rather than a hunch. It is a neat piece of product thinking - designing for the anxiety the customer has not always put into words.
Edexia stores data encrypted and de-identified on Australian servers, does not train its models on customer data, and holds SOC 2 Type II, ISO 27001 and ST4S accreditations. For a school leader signing off on student data, that list is not fine print - it is the decision.
05 / THE MARKETFree for teachers, sold to schools
Edexia's distribution runs against the usual edtech grain. Most tools sell to administrators and hope teachers fall in line. Edexia is free for individual teachers and students, which lets it grow from the classroom up, then converts that ground-level adoption into paid institutional contracts. It is purpose-built for secondary English across five curricula - the International Baccalaureate, VCE, HSC, QCE and WACE - a deliberately fragmented set of standards that a generic grader tends to fumble.
That focus is the differentiator. A wave of AI grading tools has appeared - CoGrader, EssayGrader, and older names like Turnitin and Crowdmark - and most of them apply one rubric to everyone. Edexia's premise is that the rubric is personal: your marking, not the machine's average of a million other markers. That is the line between a tool teachers tolerate and one they adopt. It is now live in 40 to 50-plus schools, among them Brisbane Grammar, St Margaret's, Gregory Terrace, Cranbrook and Helena College.
The business model reflects that same patience. Edexia gives the product away to teachers and students, then sells full deployments - with the security, administration and moderation that a school needs - as institutional contracts. It is a bottom-up motion in a market usually sold top-down, and it means the people who use the product every day are the ones who ask for the budget. Recognition has followed the adoption: features at TEDx, a listing in the ISTE EdTech Index, and coverage on 9 News.
06 / THE MONEYThe $2 million they said no to
When it came time to raise, Gibbon and Wang did something founders rarely brag about: they turned money down. Offered around $2 million each from two leading Australian venture firms, they instead took Y Combinator's standard $500,000 and a spot in the Winter 2025 batch, betting that the right launchpad mattered more than the bigger cheque. Total reported backing has since climbed past US$4 million - with more than A$6.5 million in the company overall - and the cap table includes the co-founders of ClassDojo, operators who have already scaled edtech and know a classroom when they see one.
The goal is to have a positive impact on the education system - to let educators train the tool to match their own marking styles. Daniel Gibbon, Co-Founder & CEO
Who actually uses it says a lot. Edexia's early adopters are English departments at some of Australia's more demanding schools - the kind of places where marks are moderated, defended, and occasionally appealed. Winning trust there is harder than winning it in a low-stakes classroom, and it is the reason the accuracy numbers were run on 579 real essays rather than a curated sample. The founders seem to understand that in education, credibility compounds slowly and disappears fast.
The founders relocated to San Francisco for YC, but the product's centre of gravity is still the Australian staffroom - which may be exactly why it works. Edexia was not designed to impress a demo audience. It was designed to survive a Sunday afternoon of marking, and to give some of that Sunday back.