Research, product updates, and thinking
From the ren team.
Teach the AI your house style
Every department marks differently. Ren learns your standard through two complementary mechanisms: a continuous style guide built from teacher edits, and an upfront calibration pass before batch grading.
JustinConfidence scores are not a safety net
We tested the AI's own confidence score against tens of thousands of production marks. It has almost no range, and it points the wrong way. Here is what we use instead.
Justin426 reports at university scale: our IS1108 pilot with NUS Computing
Our largest university pilot to date. Ren marked individual reports for all 426 students in NUS Computing's IS1108, grounding feedback directly in course concepts.
NatashaRen and MENDAKI: closing the feedback gap for A-Level students
Ren is partnering with Yayasan MENDAKI to support Junior College students preparing for their A-Levels with faster, more detailed assignment feedback.
Two wins, 5,000 students, four months: Ren's year at N-House pitch night
Ren won NUS Enterprise's N-House pitch night for the second consecutive year. The real story is the 12 institutional pilots, 50 teachers, and 5,000 students using the platform every week.
An AI that scores perfectly can still be useless
Measuring good essay feedback is much harder than checking a numerical score. Here is how we rebuilt our benchmark methodology to evaluate feedback quality directly.
Built by people who use it
We run our own digital academy on the same software we sell to schools. It is the fastest way to find out what works and what falls apart.
Why AI marking still needs human oversight
Language models can draft accurate feedback in seconds. Without teacher oversight, they amplify subtle errors and miss classroom context. Here is how we design for teacher-in-the-loop verification.
Better feedback loops in education
How a structured tag system turns one-off margin notes into clear diagnostic data for teachers and departments.
Our benchmark results: how Ren compares to human markers
We tested Ren against experienced teachers on real student papers. Here is what the data showed.