About ACEpath: From Answer Engines to Cognitive Companions

ACEpath grew out of a candid realization after a decade on the frontlines of educational technology. Our founder, Hui, previously held product operations and executive leadership roles at TAL Education Group (NYSE: TAL) and has spent years building tools used by millions of students.

Along the way, our field research kept running into the same troubling pattern: more than 80% of students who use camera-based homework apps use them simply to copy down the final number. They aren’t trying to learn; they are just trying to get the assignment out of the way.

Even students who wanted to check their work had a hard time doing it. They might write out ten lines of calculus on paper, then open an app and scroll through a long explanation just to figure out where they went wrong. It was slow, tedious, and drained whatever curiosity they had left.

“We spent ten years building search tools that gave kids answers in three seconds. But we never actually built a teacher. We made it effortless to finish an assignment, but in the process, we made it harder for kids to trust their own thinking. We founded ACEpath to reverse that trajectory.”

— Hui, Founder & CEO of ACEpath

Every evening around 8:30 PM, the same scene plays out in homes across North America and around the world. Dinner wraps up. Backpacks hit the kitchen table. And within thirty minutes, a quiet evening turns into an argument over calculus or physics.

Parents want to help, but after a long day they may not have touched anything about STEM in twenty years. Teenagers also want to figure things out on their own; having a parent lean over their shoulder can feel like micromanagement. Parents end up caught between stepping in too much and not knowing how to help, while students can feel more pressured than supported.

Private tutoring could bridge that gap, but good tutors are expensive—often $80 to $150 an hour—and mostly live near university campuses, far from the suburban and rural neighborhoods where families actually live.

We designed ACEpath to take that tension out of the house. By giving students an objective, infinitely patient feedback loop right at their desk, parents can stop acting as homework police and go back to being their kid’s biggest supporter.

Anyone who has ever had a great teacher knows the feeling of being listened to. A good tutor doesn’t grab your pen and start explaining everything. They watch you work, notice where you get stuck, and give you just enough help toward the next step.

We built that exact restraint into ACEpath:

• Praise what works first: Before pointing out a mistake, ACEpath highlights the steps the student got right. When kids see green checkmarks on their own scratchwork, they don’t feel like everything they did was wrong.
• No answer spoilers: ACEpath never hands over the final number upfront. It gives just enough guidance so the student experiences the click of solving it themselves.
• Protecting student agency: If the AI does the heavy lifting, the student learns nothing. Our goal is for students to understand how they got there on their own.

“Other AI only sees “right and wrong.” We see the student biting their pencil late into the night, acting as the encourager and guardian of their moments of hesitation.”

— ACEpath Core Belief

Instead of forcing students to type math into rigid input boxes, ACEpath meets them where they actually work: on pencil and paper. Our system runs across four specialized layers:

• Layer 1: Handwriting Perception: Reads messy student scratchwork with character-level spatial tracking. It handles crossed-out lines, sideways margin notes, and multi-tier fractions with over 97% step-level accuracy.
• Layer 2: Thought Reconstruction: Rebuilds the student’s internal logic across six dimensions: recognizing constraints, matching models, picking strategies, executing steps, pulling in formulas, and managing multi-body systems.
• Layer 3: Root-Cause Attribution: Separates real conceptual misunderstandings (like picking the wrong physics law) from simple clerical errors (like dropping a negative sign in line four).
• Layer 4: Adaptive Practice: Builds a continuous map of what the student knows, pairing every corrected mistake with a fresh problem that tests the exact same concept in a new context.

Most educational AI is trained on synthetic data—clean, computer-generated math that looks nothing like a real teenager’s workbook. ACEpath is built on over 100 million real handwritten student steps gathered across 3.1 billion practice logs through partner school platforms.

That real-world training shows up in student results. In pilot tracking across secondary school cohorts, an underperforming class that started 4.32 points behind the grade average ended the semester 7.06 points ahead—a net turnaround of +11.38 points on standardized exams.

For North American high schoolers, this matches how exams are actually scored. On AP Calculus and AP Physics free-response questions (FRQs), the College Board awards 80% to 90% of the score for your intermediate steps, not your final answer.

The cognitive model behind ACEpath is detailed in our research monograph, “Beyond Knowledge Concepts: A Six-Step Problem-Solving Framework for LLM-Assisted Personalized Learning.” We are actively sharing and discussing this work with peers at international educational research meetings, including AERA and AIED.

ACEpath is led by Founder & CEO Hui alongside Chief Scientist Teng Guo, Ph.D. Our core team unites veteran engineers, cognitive scientists, and curriculum specialists from leading education companies like TAL Education Group and research institutions like Purdue University and EPFL.

ACEpath operates globally through its headquarters: