Adaptive Learning Engine
A self-mutating knowledge graph that autonomously re-routes curriculum based on learner telemetry, operating on dual feedback loops.
- Author
- Radesh Govind
- Created
- July 30, 2026
Motivation
Static curricula don’t adapt to individual learners. When a student struggles with a concept, the system should detect it, diagnose the root cause, and restructure the learning path — all without human intervention. Current LMS platforms treat content as fixed sequences. This engine treats curriculum as a living, mutable graph.
Proposal
A self-optimizing backend that dynamically generates and modifies a learning curriculum. It operates on two feedback loops:
- Micro-loop — diagnoses individual failures in real-time and generates targeted remediation
- Macro-loop — structurally mutates the global curriculum based on aggregate statistical thresholds
Core Components
1. Remediation Cache
Vectorizes error profiles and caches AI-generated mitigation plans so similar struggles get instant fixes without redundant LLM calls. When a learner fails a node, their error profile is embedded and compared against cached remediations. Cache hit = instant fix. Cache miss = LLM generates a new plan, which is then cached.
2. Curriculum Mutator
When a node’s failure rate exceeds a configurable threshold (e.g., >30%), an async worker auto-generates a bridge node and mutates the DAG for all future learners. This is structural — it changes the graph itself, not just individual paths through it.
3. Tutor Dashboard
Exposes heatmaps of the curriculum graph, showing friction points, recent AI mutations, and remediation effectiveness. Tutors see which nodes are problematic and which AI-generated bridges are working.
Data Model
- Curriculum is a Directed Acyclic Graph (DAG) where nodes are learning units and edges are prerequisites
- Each node tracks: completion rate, average attempts, failure patterns, AI-generated alternatives
- Learner state is a pointer into the DAG plus a history of attempts and remediations
Alternatives Considered
- Spaced repetition (Anki-style) — optimizes review timing but doesn’t restructure content
- Adaptive testing (CAT) — adjusts difficulty but doesn’t generate new content or modify curriculum structure
- Manual curriculum updates — doesn’t scale, can’t react to individual learner patterns in real-time
Decision
Pending — draft stage.