MentoroidAI — Adaptive Learning Engine

EdTech AI · ongoing · 2026 · Client: Confidential — EdTech Client

Rule-based adaptive learning engine that personalizes K-10 learning paths using real-time mastery, stress, and confidence signals.

The problem

Traditional quiz-style learning apps optimize for correct answers, not real understanding, and ignore whether a student is cognitively overloaded or emotionally disengaged — leading to burnout, false mastery signals, and broken learning paths.

The solution

A 7-layer rule-based architecture where an Adaptive Engine performs real-time mastery estimation and node selection, a Stress & Behavioral Model tracks cognitive load and emotional tilt, a Confidence Engine gates advancement on signal reliability, and a Fail-Safe Governance layer acts as final authority over every decision via a 7-guard priority pipeline.

Key features

Real-time mastery tracking with adaptive learning-rate decay per concept
Dual-vector stress model (cognitive load + emotional tilt) driving 4 intervention states
Confidence engine combining behavioral, interaction, and calibrated self-report signals to gate node advancement
Fail-safe governance layer with a 7-guard priority hierarchy that overrides engine output when student safety is at risk
Universal knowledge graph with prerequisite-aware recovery paths, shared across multiple programs/boards

Impact

Architecture designed for sub-200ms end-to-end decision latency, with the adaptive node decision constrained to under 50ms
Decision pipeline built for 100% test coverage on all safety guards before any engine output is trusted
Replaces opaque scoring with an explainable decision trail across mastery, confidence, and stress signals for every interaction

Tech stack

Rule-based decision engine, ML Shadow Layer (observation-only), Python, FastAPI, Node.js (Fastify gateway), React Native, Next.js, Neo4j (knowledge graph), PostgreSQL, ClickHouse (event telemetry), Redis, AWS ECS Fargate, Docker, GitHub Actions

← All work · Start a project