Adaptive AI engine using deep knowledge tracing to personalize TOEIC and standardized test preparation—delivering +176% score improvement while cutting required study time in half.
Score Improvement
Faster Prep
Pass Rate
Key Outcomes
+224 TOEIC points in 10 weeks vs. +70 with traditional non-adaptive programs
58% less study time required—42 hours vs. 100+ hours to reach target score
Deep knowledge tracing updates mastery estimates across concept dependency graph with every answer
Maximum information question selection ensures every practice question maximizes learning value
Score trajectory projection creates goal-directed motivation unavailable in generic prep programs
Riiid Labs deploys an AI adaptive testing engine using deep knowledge tracing that maintains a real-time model of each student's mastery state across 300+ granular concept nodes. The system predicts the single next question that will maximize expected score improvement for that specific learner at that moment—implementing AI-driven spaced repetition and generating personalized score trajectory forecasts. Students using the adaptive engine achieve +176% score gains vs. non-adaptive study plans in the same time period, because every question serves a precise learning purpose.
Riiid Labs is an AI edtech company whose Santa AI engine powers personalized test preparation for high-stakes exams including TOEIC, SAT, and professional certification tests. Riiid's platform is particularly significant in South Korea, Japan, and Southeast Asia where standardized test scores have profound impact on career trajectories and university admission.
Traditional test prep forced students through identical content sequences regardless of their individual knowledge gaps. Advanced students wasted time on mastered concepts; struggling students hit advanced material before foundational gaps were addressed. Score plateaus were common, motivation dropped, and pass rates remained frustratingly low despite massive time investment.
34%
Content Relevance Rate
Only 34% of questions studied were at the right difficulty level for each student—the rest were either too easy or too hard to drive improvement.
100+ hrs
Average Study Time Required
Students in traditional programs spent 100+ hours to achieve target score improvements that adaptive AI achieves in 42 hours.
+70 pts
Typical Score Improvement
Non-adaptive programs delivered +70 points on TOEIC vs. +224 points with Riiid's AI-powered adaptive path.
AGIX Technologies built an AI adaptive testing engine using deep knowledge tracing (DKT) that models each student's mastery probability across 300+ concept nodes and selects the next question using maximum information theory—always choosing the question that will provide the most information about the student's true ability.
Deep Knowledge Tracing (DKT)
LSTM-based knowledge state model that tracks mastery probability for 300+ concept nodes per student, updating continuously with every answered question.
Maximum Information Question Selection
Each next question is selected by calculating the expected information gain across all available questions—choosing the item that maximally reduces uncertainty about the student's true ability.
Score Trajectory Forecasting
Personalized score prediction model that forecasts expected score at target exam date based on current mastery state, study pace, and historical learning velocity data from similar learner profiles.
Spaced Repetition Optimization
AI-driven review scheduling that times review of each concept at the predicted optimal moment before memory decay—eliminating both too-early (wasteful) and too-late (forgotten) review.
Concept Dependency Mapping
Prerequisite relationship map for all 300+ concepts ensures foundational knowledge is solidified before dependent skills are introduced, preventing frustration from premature advanced exposure.
Study Efficiency Dashboard
Student-facing analytics showing time-to-target-score, most valuable concepts to study next, and mastery progression across all skill areas—creating accountability and motivation.
TOEIC Score Gain
Adaptive AI path in 10 weeks vs. +70 points on the same traditional curriculum
Study Hours Required
42 hours to target score vs. 100+ hours with non-adaptive study plans
Content Relevance Rate
vs. 34% with traditional plans—nearly every question is at the right difficulty level
Pass Rate Improvement
Higher proportion of students reaching target score cutoffs for career advancement
"Traditional test prep is like using a map without knowing where you are. Santa AI gives every student GPS-level precision—it knows exactly which concepts need work and the fastest route to mastery. That's why our students gain 3x more points in half the time."
VP of Test Innovation
Riiid Labs
Establish the student's starting knowledge state across all concept nodes
A short adaptive diagnostic (25-40 questions) maps the student's initial mastery probabilities across all 300+ concept nodes. Questions are selected to maximize information gain—efficiently establishing baseline rather than exhaustive testing of every concept.
Every Question Has a Purpose
Maximum information question selection eliminates wasted practice—every question is chosen because it provides the most information about the student's current ability, not because it's next in a sequence.
Concept Dependencies Respected
The concept dependency graph ensures students master foundational skills before attempting dependent concepts, eliminating the frustration of hitting advanced material before prerequisites are solid.
Score Visibility Creates Accountability
Showing students their projected score and time-to-target creates goal-directed motivation that generic progress indicators cannot match—students can see exactly how today's study session moves the needle.
Forgetting Curve Integration
Spaced repetition that adapts to each student's personal retention rate (not a fixed interval formula) dramatically reduces forgotten material—improving both efficiency and retention.
Item Bank Size and Quality
The system's effectiveness scales with item bank size and calibration quality. Riiid's large, well-calibrated item bank provides enough items to support maximum information selection without repetition.
Market Context: High-Stakes Exams
In markets where TOEIC scores determine career advancement, students are highly motivated to optimize their prep. The platform's efficiency promise resonated strongly in time-constrained professional markets.
Every AI system has constraints. Here's what to know before building something similar.
Requires Large, Calibrated Item Bank
Maximum information question selection only works with a large, well-calibrated item bank. The system cannot operate effectively with fewer than ~1,000 items per subject.
Cold Start Diagnostic Has Uncertainty
The initial diagnostic has higher uncertainty than later predictions. Students who perform unusually well or poorly on diagnostic questions can cause the model to start from a misestimated baseline.
Subject-Specific Knowledge Graphs Required
Each test subject requires a custom concept dependency graph and item bank mapping. Launching new subjects requires significant curriculum mapping and item tagging investment.
Doesn't Replace Test Strategy Training
The adaptive engine optimizes knowledge mastery but doesn't cover test-taking strategy—time management, elimination techniques, and format-specific skills require separate instruction.
Explore the services, industry solutions, and intelligence types that power this system.
Common questions about building adaptive ai test preparation systems like the one deployed at Riiid Labs.