Knewton : Real-Time Adaptive Learning Powered by AI Feedback Loops

Education / EdTech

Knewton – Adaptive Learning with Real-Time AI Feedback

We enhanced Knewton’s platform with an AI layer that makes every lesson adaptive and responsive to learner needs. The system dynamically adjusts quizzes, exercises, and content flow based on student comprehension and pace. By combining user activity data with cognitive learning models, the AI identifies knowledge gaps instantly and redirects focus accordingly. It also provides instructors with dashboards that show class-wide learning trends and risk zones. This guarantees that every student keeps up in dynamic digital classrooms. Knewton now provides intelligent and engaging learning experiences on a large scale.

Project Overview

Client: Knewton (Leading adaptive learning platform acquired by Wiley)
Challenge: Rigid content delivery limited student engagement and failed to adapt to real-time learning progress
Goal: Enhance the platform with AI to:
- Dynamically personalize lesson paths based on comprehension and pace
- Spot areas of weakness and adapt assessments in response.
- Give instructors real-time insights into class-wide performance trends
Team: 8 (3 AI Engineers, 2 EdTech Strategists, 2 Frontend Devs, 1 Data Analyst)
Timeline: 5-months path to full deployment
(Knewton needed to move from fixed paths to an AI-driven adaptive classroom.)

By leveraging GenX’s AI, Knewton evolved from responding to changes to predicting them. Our platform now makes adjustments in real-time, rather than retroactively.

VP of Product, Knewton

The Challenge

Critical Pain Points:

Students often advanced without fully mastering earlier concepts
Teachers lacked visibility into who was falling behind in real time
Content failed to adjust to different learning speeds and retention levels

Technical Hurdles:

Building micro-adaptivity into exercises without redesigning core curriculum
Modeling comprehension from clickstream, quiz attempts, and time-on-task
Ensuring the AI could explain adaptation decisions to both students and instructors

Tech Stack

Component	Technologies
Learning Models	TensorFlow, BKT, PyTorch, HuggingFace Transformers
Data Processing	Google Dataflow, BigQuery, Firebase Firestore
Adaptive Content Engine	FAISS, Sentence-BERT, Bloom-Aligned Rule Layer
Backend + API Layer	FastAPI, Node.js, PostgreSQL
Dashboards & Notifications	React, D3.js, Firebase Cloud Messaging
Monitoring & Privacy	Sentry, Mixpanel, COPPA/GDPR Compliance Modules

Key Innovations

Knewton’s system dynamically adapted content flow based on real-time student understanding. Knowledge gaps were swiftly identified by AI, which allowed for immediate adjustments, while educators used dashboards to track and modify instruction as needed

Micro-Adaptive Content Flows

Adjusted lesson paths and quiz sets in real time as students progressed

Result: 34% increase in content engagement duration

Comprehension-Driven Alerts for Instructors

Detected students likely to underperform based on early session signals

Result: 22% drop in end-of-module failure rates

Dynamic Remediation Engine

Delivered extra practice and simpler content when gaps were detected

Result: 29% improvement in concept mastery scores after remediation

Our AI/ML Architecture

Core Models

Comprehension Detection Engine
- Built on Bayesian Knowledge Tracing (BKT) + RNN hybrids
- Scores real-time understanding to drive personalized lesson adjustments
Exercise & Quiz Generator
- A system that rearranges exercises using semantic similarity.
- Aligns alternate questions with Bloom’s taxonomy levels
Instructor Insight Module
- Visualizes class-wide heatmaps of struggling topics
- Predicts drop-off and disengagement using engagement decay curves

Data Pipeline

Sources
- Real-time quiz submissions, interaction history, and learning session data.
- Student profile history, prior mastery levels, and subject-specific difficulty matrices