CogniFlow
Team project · BloodCoded
Agentic AIKnowledge TracingAdaptive Learning
Problem Statement
When a student fails the same exercise twice, most AI tutors hand them an easier version of it — but the real gap is often a missing prerequisite, so they fail again for the same invisible reason.
Solution
Built an agentic tutor that traces repeated failure to its cause: it walks a prerequisite graph using live Bayesian mastery estimates, reassigns its own objective to the weakest unmastered skill, teaches it, then returns to the original goal.
Key Features
- ▹Prerequisite-aware redirects over a skill dependency graph
- ▹Bayesian Knowledge Tracing for live mastery estimates
- ▹LLM decisions validated by a deterministic policy guard
- ▹Infrastructure faults can never move a student's mastery
- ▹19% fewer attempts to mastery across 80 simulated students